Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Inpainting","local":"inpainting","sections":[{"title":"Tạo một hình ảnh mặt nạ","local":"create-a-mask-image","sections":[{"title":"Làm mờ mặt nạ","local":"mask-blur","sections":[],"depth":3}],"depth":2},{"title":"Các mô hình phổ biến","local":"popular-models","sections":[{"title":"Stable Diffusion Inpainting","local":"stable-diffusion-inpainting","sections":[],"depth":3},{"title":"Stable Diffusion XL (SDXL) Inpainting","local":"stable-diffusion-xl-sdxl-inpainting","sections":[],"depth":3},{"title":"Kandinsky 2.2 Inpainting","local":"kandinsky-22-inpainting","sections":[],"depth":3}],"depth":2},{"title":"Các checkpoint không dành riêng cho inpainting","local":"non-inpaint-specific-checkpoints","sections":[],"depth":2},{"title":"Cấu hình tham số pipeline","local":"configure-pipeline-parameters","sections":[{"title":"Độ mạnh","local":"strength","sections":[],"depth":3},{"title":"Thang độ dẫn","local":"guidance-scale","sections":[],"depth":3},{"title":"Negative prompt","local":"negative-prompt","sections":[],"depth":3},{"title":"Cắt theo mặt nạ đệm","local":"padding-mask-crop","sections":[],"depth":3}],"depth":2},{"title":"Các pipeline inpainting được nối tiếp","local":"chained-inpainting-pipelines","sections":[{"title":"Từ văn bản sang hình ảnh sang inpaint","local":"text-to-image-to-inpaint","sections":[],"depth":3},{"title":"Inpaint-to-image-to-image","local":"inpaint-to-image-to-image","sections":[],"depth":3}],"depth":2},{"title":"Kiểm soát tạo ảnh","local":"control-image-generation","sections":[{"title":"Trọng số prompt","local":"prompt-weighting","sections":[],"depth":3},{"title":"ControlNet","local":"controlnet","sections":[],"depth":3}],"depth":2},{"title":"Tối ưu","local":"optimize","sections":[],"depth":2}],"depth":1}"/> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/start.B6roO9Zr.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/CjYFLaCl.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/Cc4T7sti.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/C8F28shN.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/BTnLXsn-.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DnVIJfC5.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/nodes/0.CmHT0Edy.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/C7Ky3Ped.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/nodes/15.ByMW7QSd.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/D2YP2_0L.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/DafAQkiX.js" rel="modulepreload"> | |
| <link href="/docs/diffusers/main/en/_app/immutable/chunks/i2teRydZ.js" rel="modulepreload"> | |
| <!--1vpuhl9--><meta name="hf:doc:metadata" content="{"title":"Inpainting","local":"inpainting","sections":[{"title":"Create a mask image","local":"create-a-mask-image","sections":[{"title":"Mask blur","local":"mask-blur","sections":[],"depth":3}],"depth":2},{"title":"Popular models","local":"popular-models","sections":[{"title":"Stable Diffusion Inpainting","local":"stable-diffusion-inpainting","sections":[],"depth":3},{"title":"Stable Diffusion XL (SDXL) Inpainting","local":"stable-diffusion-xl-sdxl-inpainting","sections":[],"depth":3},{"title":"Kandinsky 2.2 Inpainting","local":"kandinsky-22-inpainting","sections":[],"depth":3}],"depth":2},{"title":"Non-inpaint specific checkpoints","local":"non-inpaint-specific-checkpoints","sections":[],"depth":2},{"title":"Configure pipeline parameters","local":"configure-pipeline-parameters","sections":[{"title":"Strength","local":"strength","sections":[],"depth":3},{"title":"Guidance scale","local":"guidance-scale","sections":[],"depth":3},{"title":"Negative prompt","local":"negative-prompt","sections":[],"depth":3},{"title":"Padding mask crop","local":"padding-mask-crop","sections":[],"depth":3}],"depth":2},{"title":"Chained inpainting pipelines","local":"chained-inpainting-pipelines","sections":[{"title":"Text-to-image-to-inpaint","local":"text-to-image-to-inpaint","sections":[],"depth":3},{"title":"Inpaint-to-image-to-image","local":"inpaint-to-image-to-image","sections":[],"depth":3}],"depth":2},{"title":"Control image generation","local":"control-image-generation","sections":[{"title":"Prompt weighting","local":"prompt-weighting","sections":[],"depth":3},{"title":"ControlNet","local":"controlnet","sections":[],"depth":3}],"depth":2},{"title":"Optimize","local":"optimize","sections":[],"depth":2}],"depth":1}"/><!----> | |
| <link href="/docs/diffusers/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;"><div dir="ltr" class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----></span> <span dir="auto">Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <div class="flex space-x-1 " style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;" dir="ltr"><!----> <!--[1--><div class="relative colab-dropdown "><button class=" " type="button"><!--[0--><!--[-1--><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"/><!--]--><!----><!--]--></button> <!--[-1--><!--]--></div><!--]--> <!--[1--><div class="relative colab-dropdown "><button class=" " type="button"><!--[0--><!--[-1--><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"/><!--]--><!----><!--]--></button> <!--[-1--><!--]--></div><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="inpainting" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#inpainting"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Inpainting<!----></span></h1><!--]--><!----> <p>Inpainting thay thế hoặc chỉnh sửa các khu vực cụ thể của một hình ảnh. Điều này biến nó thành một công cụ hữu ích cho việc phục hồi hình ảnh như loại bỏ các khuyết điểm và hiện tượng nhiễu, hoặc thậm chí thay thế một khu vực của hình ảnh bằng một thứ gì đó hoàn toàn mới. Inpainting dựa vào một mask để xác định các vùng của hình ảnh cần được lấp đầy; khu vực cần inpaint được biểu diễn bằng các pixel màu trắng và khu vực cần giữ lại được biểu diễn bằng các pixel màu đen. Các pixel màu trắng được lấp đầy bởi prompt.</p> <p>Với 🤗 Diffusers, đây là cách bạn có thể thực hiện inpainting:</p> <ol><li>Tải checkpoint inpainting bằng lớp <a href="/docs/diffusers/main/vi/api/pipelines/auto_pipeline#diffusers.AutoPipelineForInpainting">AutoPipelineForInpainting</a>. Lớp này sẽ tự động phát hiện lớp pipeline phù hợp để tải dựa trên checkpoint:</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"kandinsky-community/kandinsky-2-2-decoder-inpaint"</span>, dtype=torch.float16 | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention()<!----></pre></div><!----> <blockquote class="tip"><p>Bạn sẽ nhận thấy trong suốt hướng dẫn, chúng tôi sử dụng <a href="/docs/diffusers/main/vi/api/pipelines/overview#diffusers.DiffusionPipeline.enable_model_cpu_offload">enable_model_cpu_offload()</a> và <a href="/docs/diffusers/main/vi/api/pipelines/overview#diffusers.DiffusionPipeline.enable_xformers_memory_efficient_attention">enable_xformers_memory_efficient_attention()</a> để tiết kiệm bộ nhớ và tăng tốc độ suy luận. Nếu bạn đang sử dụng PyTorch 2.0, không cần thiết phải gọi <a href="/docs/diffusers/main/vi/api/pipelines/overview#diffusers.DiffusionPipeline.enable_xformers_memory_efficient_attention">enable_xformers_memory_efficient_attention()</a> trên pipeline của bạn vì nó đã sử dụng <a href="../optimization/fp16#scaled-dot-product-attention">scaled-dot product attention</a> bản địa của PyTorch 2.0.</p></blockquote> <ol start="2"><li>Tải ảnh nền và ảnh mặt nạ:</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>)<!----></pre></div><!----> <ol start="3"><li>Tạo một prompt để inpaint hình ảnh và truyền nó vào pipeline cùng với hình ảnh gốc và hình ảnh mask:</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->prompt = <span class="hljs-string">"a black cat with glowing eyes, cute, adorable, disney, pixar, highly detailed, 8k"</span> | |
| negative_prompt = <span class="hljs-string">"bad anatomy, deformed, ugly, disfigured"</span> | |
| image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex gap-4"><div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ảnh cơ sở</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">mask image</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-cat.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">hình ảnh được tạo</figcaption></div></div> <!--[1--><h2 class="relative group"><a id="create-a-mask-image" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#create-a-mask-image"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Tạo một hình ảnh mặt nạ<!----></span></h2><!--]--><!----> <p>Trong suốt hướng dẫn này, hình ảnh mask được cung cấp trong tất cả các ví dụ mã để tiện lợi. Bạn có thể inpaint trên các hình ảnh của riêng mình, nhưng sẽ cần tạo một hình ảnh mask cho nó. Sử dụng Space bên dưới để dễ dàng tạo một hình ảnh mask.</p> <p>Tải lên một hình ảnh nền để inpaint và sử dụng công cụ phác thảo để vẽ mặt nạ. Khi hoàn tất, nhấp vào <strong>Run</strong> để tạo và tải xuống hình ảnh mặt nạ.</p> <iframe src="https://stevhliu-inpaint-mask-maker.hf.space" frameborder="0" width="850" height="450"></iframe> <!--[2--><h3 class="relative group"><a id="mask-blur" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#mask-blur"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Làm mờ mặt nạ<!----></span></h3><!--]--><!----> <p>Phương thức <a href="/docs/diffusers/main/vi/api/image_processor#diffusers.VaeImageProcessor.blur">blur()</a> cung cấp tùy chọn về cách trộn lẫn giữa ảnh gốc và khu vực inpaint. Mức độ làm mờ được xác định bởi tham số <code>blur_factor</code>. Việc tăng <code>blur_factor</code> sẽ tăng lượng làm mờ được áp dụng lên các cạnh của mask, làm mềm sự chuyển tiếp giữa ảnh gốc và khu vực inpaint. Một <code>blur_factor</code> thấp hoặc bằng không sẽ giữ cho các cạnh của mask sắc nét hơn.</p> <p>Để sử dụng tính năng này, hãy tạo một mặt nạ mờ bằng trình xử lý ảnh.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image | |
| <span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image | |
| pipeline = AutoPipelineForInpainting.from_pretrained(<span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, dtype=torch.float16).to(<span class="hljs-string">'cuda'</span>) <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| mask = load_image(<span class="hljs-string">"https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/seashore_mask.png"</span>) | |
| blurred_mask = pipeline.mask_processor.blur(mask, blur_factor=<span class="hljs-number">33</span>) | |
| blurred_mask<!----></pre></div><!----> <div class="flex gap-4"><div><img class="rounded-xl" src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/seashore_mask.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">mask không có làm mờ</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/mask_blurred.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">mask có áp dụng hiệu ứng làm mờ</figcaption></div></div> <!--[1--><h2 class="relative group"><a id="popular-models" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#popular-models"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Các mô hình phổ biến<!----></span></h2><!--]--><!----> <p><a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-inpainting" rel="nofollow">Stable Diffusion Inpainting</a>, <a href="https://huggingface.co/diffusers/stable-diffusion-xl-1.0-inpainting-0.1" rel="nofollow">Stable Diffusion XL (SDXL) Inpainting</a> và <a href="https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder-inpaint" rel="nofollow">Kandinsky 2.2 Inpainting</a> là một trong những mô hình inpainting phổ biến nhất. SDXL thường tạo ra các hình ảnh có độ phân giải cao hơn so với Stable Diffusion v1.5, và Kandinsky 2.2 cũng có khả năng tạo ra các hình ảnh chất lượng cao.</p> <!--[2--><h3 class="relative group"><a id="stable-diffusion-inpainting" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#stable-diffusion-inpainting"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Stable Diffusion Inpainting<!----></span></h3><!--]--><!----> <p>Stable Diffusion Inpainting là một latent diffusion model được finetune trên các ảnh 512x512 cho tác vụ inpainting. Đây là một điểm khởi đầu tốt vì nó tương đối nhanh và tạo ra các ảnh có chất lượng tốt. Để sử dụng model này cho inpainting, bạn cần truyền một prompt, ảnh base và ảnh mask vào pipeline:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| generator = torch.Generator(<span class="hljs-string">"cuda"</span>).manual_seed(<span class="hljs-number">92</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="stable-diffusion-xl-sdxl-inpainting" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#stable-diffusion-xl-sdxl-inpainting"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Stable Diffusion XL (SDXL) Inpainting<!----></span></h3><!--]--><!----> <p>SDXL là phiên bản lớn hơn và mạnh hơn của Stable Diffusion v1.5. Mô hình này có thể tuân theo quy trình mô hình hai giai đoạn (mặc dù mỗi mô hình cũng có thể được sử dụng riêng lẻ); mô hình cơ sở tạo ra một hình ảnh, và mô hình tinh chỉnh nhận hình ảnh đó để nâng cao thêm chi tiết và chất lượng. Hãy xem hướng dẫn <a href="../api/pipelines/stable_diffusion/stable_diffusion_xl">SDXL</a> để có hướng dẫn toàn diện hơn về cách sử dụng SDXL và cấu hình các tham số của nó.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"diffusers/stable-diffusion-xl-1.0-inpainting-0.1"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| generator = torch.Generator(<span class="hljs-string">"cuda"</span>).manual_seed(<span class="hljs-number">92</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="kandinsky-22-inpainting" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#kandinsky-22-inpainting"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Kandinsky 2.2 Inpainting<!----></span></h3><!--]--><!----> <p>Họ model Kandinsky tương tự SDXL vì cũng sử dụng hai model; model image prior tạo ra image embeddings, và model diffusion tạo hình ảnh từ chúng. Bạn có thể nạp riêng model image prior và model diffusion, nhưng cách dễ nhất để sử dụng Kandinsky 2.2 là nạp nó vào lớp <a href="/docs/diffusers/main/vi/api/pipelines/auto_pipeline#diffusers.AutoPipelineForInpainting">AutoPipelineForInpainting</a>, lớp này sử dụng <a href="/docs/diffusers/main/vi/api/pipelines/kandinsky_v22#diffusers.KandinskyV22InpaintCombinedPipeline">KandinskyV22InpaintCombinedPipeline</a> bên trong.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"kandinsky-community/kandinsky-2-2-decoder-inpaint"</span>, dtype=torch.float16 | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| generator = torch.Generator(<span class="hljs-string">"cuda"</span>).manual_seed(<span class="hljs-number">92</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ảnh cơ sở</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-sdv1.5.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion Inpainting</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-sdxl.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion XL Inpainting</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-kandinsky.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">Kandinsky 2.2 Inpainting</figcaption></div></div> <!--[1--><h2 class="relative group"><a id="non-inpaint-specific-checkpoints" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#non-inpaint-specific-checkpoints"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Các checkpoint không dành riêng cho inpainting<!----></span></h2><!--]--><!----> <p>Cho đến nay, hướng dẫn này đã sử dụng các checkpoint dành riêng cho inpaint như <a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-inpainting" rel="nofollow">stable-diffusion-v1-5/stable-diffusion-inpainting</a>. Nhưng bạn cũng có thể sử dụng các checkpoint thông thường như <a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5" rel="nofollow">stable-diffusion-v1-5/stable-diffusion-v1-5</a>. Hãy so sánh kết quả của hai checkpoint này.</p> <p>Hình ảnh bên trái được tạo từ một checkpoint thông thường, còn hình ảnh bên phải được tạo từ một checkpoint inpaint. Bạn sẽ ngay lập tức nhận thấy hình ảnh bên trái không sạch sẽ bằng, và vẫn có thể nhìn thấy đường viền của khu vực mà mô hình cần inpaint. Hình ảnh bên phải sạch sẽ hơn nhiều và khu vực được inpaint trông tự nhiên hơn.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><!--[--><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">stable-diffusion-v1-5/stable-diffusion-v1-5</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">stable-diffusion-v1-5/stable-diffusion-inpainting</div><!--]--></div> <div class="language-select"><div><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| generator = torch.Generator(<span class="hljs-string">"cuda"</span>).manual_seed(<span class="hljs-number">92</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----></div><!----> <div hidden=""><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| generator = torch.Generator(<span class="hljs-string">"cuda"</span>).manual_seed(<span class="hljs-number">92</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----></div><!----><!----></div><!----> <div class="flex gap-4"><div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/non-inpaint-specific.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">stable-diffusion-v1-5/stable-diffusion-v1-5</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-specific.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">stable-diffusion-v1-5/stable-diffusion-inpainting</figcaption></div></div> <p>Tuy nhiên, đối với các tác vụ cơ bản hơn như xóa một đối tượng khỏi hình ảnh (ví dụ như những tảng đá trên đường), một checkpoint thông thường cho kết quả khá tốt. Không có sự khác biệt đáng kể nào giữa checkpoint thông thường và checkpoint inpaint.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><!--[--><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">stable-diffusion-v1-5/stable-diffusion-v1-5</div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">stable-diffusion-v1-5/stable-diffusion-inpaint</div><!--]--></div> <div class="language-select"><div><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/road-mask.png"</span>) | |
| image = pipeline(prompt=<span class="hljs-string">"road"</span>, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----></div><!----> <div hidden=""><div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/road-mask.png"</span>) | |
| image = pipeline(prompt=<span class="hljs-string">"road"</span>, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----></div><!----><!----></div><!----> <div class="flex gap-4"><div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/regular-inpaint-basic.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">stable-diffusion-v1-5/stable-diffusion-v1-5</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/specific-inpaint-basic.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">stable-diffusion-v1-5/stable-diffusion-inpainting</figcaption></div></div> <p>Nhược điểm khi sử dụng checkpoint không chuyên về inpaint là chất lượng tổng thể của hình ảnh có thể thấp hơn, nhưng nó thường có xu hướng giữ nguyên vùng mask (đó là lý do bạn có thể thấy đường viền của mask). Các checkpoint chuyên về inpaint được huấn luyện có chủ đích để tạo ra các hình ảnh inpaint có chất lượng cao hơn, bao gồm việc tạo ra sự chuyển tiếp tự nhiên hơn giữa vùng bị mask và vùng không bị mask. Do đó, các checkpoint này có khả năng cao hơn trong việc thay đổi vùng không bị mask của bạn.</p> <p>Nếu việc giữ nguyên vùng không bị che (unmasked) là quan trọng đối với tác vụ của bạn, bạn có thể sử dụng phương thức <a href="/docs/diffusers/main/vi/api/image_processor#diffusers.VaeImageProcessor.apply_overlay">VaeImageProcessor.apply_overlay()</a> để buộc vùng không bị che của hình ảnh giữ nguyên, với cái giá là một số chuyển tiếp kém tự nhiên hơn giữa vùng bị che và vùng không bị che.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> PIL | |
| <span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np | |
| <span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| device = <span class="hljs-string">"cuda"</span> <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, | |
| dtype=torch.float16, | |
| variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline = pipeline.to(device) | |
| img_url = <span class="hljs-string">"https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"</span> | |
| mask_url = <span class="hljs-string">"https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"</span> | |
| init_image = load_image(img_url).resize((<span class="hljs-number">512</span>, <span class="hljs-number">512</span>)) | |
| mask_image = load_image(mask_url).resize((<span class="hljs-number">512</span>, <span class="hljs-number">512</span>)) | |
| prompt = <span class="hljs-string">"Face of a yellow cat, high resolution, sitting on a park bench"</span> | |
| repainted_image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| repainted_image.save(<span class="hljs-string">"repainted_image.png"</span>) | |
| unmasked_unchanged_image = pipeline.image_processor.apply_overlay(mask_image, init_image, repainted_image) | |
| unmasked_unchanged_image.save(<span class="hljs-string">"force_unmasked_unchanged.png"</span>) | |
| make_image_grid([init_image, mask_image, repainted_image, unmasked_unchanged_image], rows=<span class="hljs-number">2</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="configure-pipeline-parameters" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#configure-pipeline-parameters"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Cấu hình tham số pipeline<!----></span></h2><!--]--><!----> <p>Đặc điểm của hình ảnh - như chất lượng và “sáng tạo” - phụ thuộc vào các tham số của pipeline. Việc hiểu rõ các tham số này hoạt động như thế nào là rất quan trọng để đạt được kết quả mong muốn. Hãy cùng xem xét các tham số quan trọng nhất và cách việc thay đổi chúng ảnh hưởng đến đầu ra.</p> <!--[2--><h3 class="relative group"><a id="strength" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#strength"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Độ mạnh<!----></span></h3><!--]--><!----> <p><code>strength</code> là thước đo mức độ nhiễu được thêm vào ảnh gốc, ảnh hưởng đến mức độ tương đồng giữa ảnh đầu ra và ảnh gốc.</p> <ul><li>📈 giá trị <code>strength</code> cao hơn có nghĩa là nhiều nhiễu hơn được thêm vào hình ảnh và quá trình khử nhiễu mất nhiều thời gian hơn, nhưng bạn sẽ nhận được các hình ảnh có chất lượng cao hơn và khác biệt hơn so với hình ảnh gốc</li> <li>📉 giá trị <code>strength</code> thấp có nghĩa là ít nhiễu được thêm vào hình ảnh hơn và quá trình khử nhiễu diễn ra nhanh hơn, nhưng chất lượng hình ảnh có thể không cao và hình ảnh tạo ra sẽ giống với hình ảnh gốc hơn</li></ul> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, strength=<span class="hljs-number">0.6</span>).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-0.6.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.6</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-0.8.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.8</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-1.0.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 1.0</figcaption></div></div> <!--[2--><h3 class="relative group"><a id="guidance-scale" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#guidance-scale"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Thang độ dẫn<!----></span></h3><!--]--><!----> <p><code>guidance_scale</code> ảnh hưởng đến mức độ khớp giữa prompt văn bản và hình ảnh được tạo ra.</p> <ul><li>📈 giá trị <code>guidance_scale</code> cao có nghĩa là prompt và hình ảnh được tạo ra khớp sát nhau, do đó kết quả là một diễn giải chặt chẽ hơn của prompt</li> <li>📉 giá trị <code>guidance_scale</code> thấp có nghĩa là prompt và hình ảnh được tạo ra ít liên kết chặt chẽ hơn, do đó kết quả có thể đa dạng hơn so với prompt</li></ul> <p>Bạn có thể sử dụng <code>strength</code> và <code>guidance_scale</code> cùng nhau để kiểm soát tốt hơn mức độ biểu cảm của mô hình. Ví dụ, sự kết hợp giữa giá trị <code>strength</code> và <code>guidance_scale</code> cao sẽ mang lại cho mô hình sự tự do sáng tạo lớn nhất.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, guidance_scale=<span class="hljs-number">2.5</span>).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-2.5.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 2.5</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-7.5.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 7.5</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-12.5.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 12.5</figcaption></div></div> <!--[2--><h3 class="relative group"><a id="negative-prompt" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#negative-prompt"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Negative prompt<!----></span></h3><!--]--><!----> <p>Prompt âm đóng vai trò ngược lại với prompt; nó hướng mô hình tránh tạo ra một số yếu tố nhất định trong hình ảnh. Điều này hữu ích để nhanh chóng cải thiện chất lượng hình ảnh và ngăn mô hình tạo ra những thứ bạn không mong muốn.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| negative_prompt = <span class="hljs-string">"bad architecture, unstable, poor details, blurry"</span> | |
| image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex justify-center"><figure><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-negative.png"/> <figcaption class="text-center">negative_prompt = "bad architecture, unstable, poor details, blurry"</figcaption></figure></div> <!--[2--><h3 class="relative group"><a id="padding-mask-crop" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#padding-mask-crop"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Cắt theo mặt nạ đệm<!----></span></h3><!--]--><!----> <p>Một phương pháp để tăng chất lượng ảnh inpainting là sử dụng tham số <a href="https://huggingface.co/docs/diffusers/v0.25.0/en/api/pipelines/stable_diffusion/inpaint#diffusers.StableDiffusionInpaintPipeline.__call__.padding_mask_crop" rel="nofollow"><code>padding_mask_crop</code></a>. Khi được bật, tùy chọn này sẽ cắt vùng bị mask với một lượng padding do người dùng chỉ định và cũng cắt cùng vùng đó từ ảnh gốc. Cả ảnh và mask đều được upscale lên độ phân giải cao hơn để inpainting, sau đó được ghép đè lên ảnh gốc. Đây là cách nhanh chóng và dễ dàng để cải thiện chất lượng ảnh mà không cần sử dụng một pipeline riêng biệt như <a href="/docs/diffusers/main/vi/api/pipelines/stable_diffusion/upscale#diffusers.StableDiffusionUpscalePipeline">StableDiffusionUpscalePipeline</a>.</p> <p>Thêm tham số <code>padding_mask_crop</code> vào lệnh gọi pipeline và đặt nó thành giá trị padding mong muốn.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image | |
| <span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image | |
| generator = torch.Generator(device=<span class="hljs-string">'cuda'</span>).manual_seed(<span class="hljs-number">0</span>) <span class="hljs-comment"># or "mps", "xpu", "cpu"</span> | |
| pipeline = AutoPipelineForInpainting.from_pretrained(<span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, dtype=torch.float16).to(<span class="hljs-string">'cuda'</span>) | |
| base = load_image(<span class="hljs-string">"https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/seashore.png"</span>) | |
| mask = load_image(<span class="hljs-string">"https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/seashore_mask.png"</span>) | |
| image = pipeline(<span class="hljs-string">"boat"</span>, image=base, mask_image=mask, strength=<span class="hljs-number">0.75</span>, generator=generator, padding_mask_crop=<span class="hljs-number">32</span>).images[<span class="hljs-number">0</span>] | |
| image<!----></pre></div><!----> <div class="flex gap-4"><div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/baseline_inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ảnh inpaint mặc định</figcaption></div> <div><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/padding_mask_crop_inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">inpaint image với `padding_mask_crop` được bật</figcaption></div></div> <!--[1--><h2 class="relative group"><a id="chained-inpainting-pipelines" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#chained-inpainting-pipelines"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Các pipeline inpainting được nối tiếp<!----></span></h2><!--]--><!----> <p><a href="/docs/diffusers/main/vi/api/pipelines/auto_pipeline#diffusers.AutoPipelineForInpainting">AutoPipelineForInpainting</a> có thể được kết nối với các pipeline 🤗 Diffusers khác để chỉnh sửa đầu ra của chúng. Điều này thường hữu ích để cải thiện chất lượng đầu ra từ các pipeline khuếch tán khác của bạn, và nếu bạn đang sử dụng nhiều pipeline, việc kết nối chúng lại với nhau có thể tiết kiệm bộ nhớ hơn bằng cách giữ các đầu ra trong không gian tiềm ẩn và tái sử dụng các thành phần pipeline giống nhau.</p> <!--[2--><h3 class="relative group"><a id="text-to-image-to-inpaint" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#text-to-image-to-inpaint"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Từ văn bản sang hình ảnh sang inpaint<!----></span></h3><!--]--><!----> <p>Việc kết nối (chaining) một pipeline text-to-image và một pipeline inpainting cho phép bạn inpaint hình ảnh đã được tạo ra, và bạn không cần cung cấp một hình ảnh nền ban đầu. Điều này giúp việc chỉnh sửa các kết quả text-to-image yêu thích của bạn trở nên thuận tiện mà không cần phải tạo ra một hình ảnh hoàn toàn mới.</p> <p>Bắt đầu với pipeline text-to-image để tạo một lâu đài:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForText2Image, AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForText2Image.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span>, use_safetensors=<span class="hljs-literal">True</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| text2image = pipeline(<span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span>).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <p>Tải ảnh mask của đầu ra từ trên:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_text-chain-mask.png"</span>)<!----></pre></div><!----> <p>Và hãy inpaint vùng bị mask bằng một thác nước:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"kandinsky-community/kandinsky-2-2-decoder-inpaint"</span>, dtype=torch.float16 | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| prompt = <span class="hljs-string">"digital painting of a fantasy waterfall, cloudy"</span> | |
| image = pipeline(prompt=prompt, image=text2image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([text2image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-text-chain.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">text-to-image</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-text-chain-out.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">inpaint</figcaption></div></div> <!--[2--><h3 class="relative group"><a id="inpaint-to-image-to-image" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#inpaint-to-image-to-image"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Inpaint-to-image-to-image<!----></span></h3><!--]--><!----> <p>Bạn cũng có thể nối một pipeline inpainting trước một pipeline khác như image-to-image hoặc upscaler để cải thiện chất lượng.</p> <p>Bắt đầu bằng cách inpainting một hình ảnh:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting, AutoPipelineForImage2Image | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image_inpainting = pipeline(prompt=prompt, image=init_image, mask_image=mask_image).images[<span class="hljs-number">0</span>] | |
| <span class="hljs-comment"># resize image to 1024x1024 for SDXL</span> | |
| image_inpainting = image_inpainting.resize((<span class="hljs-number">1024</span>, <span class="hljs-number">1024</span>))<!----></pre></div><!----> <p>Bây giờ, hãy truyền ảnh vào một pipeline inpainting khác với mô hình refiner của SDXL để tăng cường chi tiết và chất lượng của ảnh:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stabilityai/stable-diffusion-xl-refiner-1.0"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| image = pipeline(prompt=prompt, image=image_inpainting, mask_image=mask_image, output_type=<span class="hljs-string">"latent"</span>).images[<span class="hljs-number">0</span>]<!----></pre></div><!----> <blockquote class="tip"><p>Việc chỉ định <code>output_type="latent"</code> trong pipeline là quan trọng để giữ tất cả các đầu ra trong không gian latent, tránh bước decode-encode không cần thiết. Điều này chỉ hoạt động nếu các pipeline được nối tiếp sử dụng cùng một VAE. Ví dụ, trong phần <a href="#text-to-image-to-inpaint">Text-to-image-to-inpaint</a>, Kandinsky 2.2 sử dụng một lớp VAE khác với mô hình Stable Diffusion nên sẽ không hoạt động. Nhưng nếu bạn sử dụng Stable Diffusion v1.5 cho cả hai pipeline, thì bạn có thể giữ mọi thứ trong không gian latent vì cả hai đều sử dụng <a href="/docs/diffusers/main/vi/api/models/autoencoderkl#diffusers.AutoencoderKL">AutoencoderKL</a>.</p></blockquote> <p>Cuối cùng, bạn có thể truyền ảnh này vào một pipeline image-to-image để hoàn thiện nó. Việc sử dụng phương thức <a href="/docs/diffusers/main/vi/api/pipelines/auto_pipeline#diffusers.AutoPipelineForImage2Image.from_pipe">from_pipe()</a> để tái sử dụng các thành phần pipeline hiện có sẽ hiệu quả hơn, và tránh việc tải lại tất cả các thành phần pipeline vào bộ nhớ một cách không cần thiết.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline = AutoPipelineForImage2Image.from_pipe(pipeline) | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| image = pipeline(prompt=prompt, image=image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image_inpainting, image], rows=<span class="hljs-number">2</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ảnh ban đầu</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-to-image-chain.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">inpaint</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-to-image-final.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">image-to-image</figcaption></div></div> <p>Image-to-image và inpainting thực chất là những tác vụ rất tương đồng. Image-to-image tạo ra một hình ảnh mới có nét giống với hình ảnh được cung cấp sẵn. Inpainting cũng làm điều tương tự, nhưng chỉ biến đổi vùng hình ảnh được xác định bởi mask, trong khi phần còn lại của hình ảnh giữ nguyên. Bạn có thể xem inpainting như một công cụ chính xác hơn để thực hiện những thay đổi cụ thể, còn image-to-image có phạm vi rộng hơn để tạo ra những thay đổi mang tính tổng quát hơn.</p> <!--[1--><h2 class="relative group"><a id="control-image-generation" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#control-image-generation"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Kiểm soát tạo ảnh<!----></span></h2><!--]--><!----> <p>Việc khiến hình ảnh trông chính xác như mong muốn là một thách thức vì quá trình khử nhiễu mang tính ngẫu nhiên. Mặc dù bạn có thể kiểm soát một số khía cạnh của quá trình tạo hình bằng cách cấu hình các tham số như <code>negative_prompt</code>, nhưng có những phương pháp tốt hơn và hiệu quả hơn để kiểm soát việc tạo hình ảnh.</p> <!--[2--><h3 class="relative group"><a id="prompt-weighting" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#prompt-weighting"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Trọng số prompt<!----></span></h3><!--]--><!----> <p>Prompt weighting cung cấp một cách định lượng để điều chỉnh mức độ biểu diễn của các khái niệm trong prompt. Bạn có thể sử dụng nó để tăng hoặc giảm độ lớn của vector nhúng văn bản cho từng khái niệm trong prompt, từ đó xác định mức độ mà mỗi khái niệm được tạo ra. Thư viện <a href="https://github.com/damian0815/compel" rel="nofollow">Compel</a> cung cấp cú pháp trực quan để điều chỉnh trọng số prompt và tạo ra các embedding. Tìm hiểu cách tạo các embedding trong hướng dẫn <a href="../using-diffusers/weighted_prompts">Prompt weighting</a>.</p> <p>Sau khi bạn đã tạo các embeddings, hãy truyền chúng vào tham số <code>prompt_embeds</code> (và <code>negative_prompt_embeds</code> nếu bạn đang sử dụng negative prompt) trong <a href="/docs/diffusers/main/vi/api/pipelines/auto_pipeline#diffusers.AutoPipelineForInpainting">AutoPipelineForInpainting</a>. Các embeddings này thay thế cho tham số <code>prompt</code>:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForInpainting | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> make_image_grid | |
| pipeline = AutoPipelineForInpainting.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, dtype=torch.float16, | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| image = pipeline(prompt_embeds=prompt_embeds, <span class="hljs-comment"># generated from Compel</span> | |
| negative_prompt_embeds=negative_prompt_embeds, <span class="hljs-comment"># generated from Compel</span> | |
| image=init_image, | |
| mask_image=mask_image | |
| ).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image], rows=<span class="hljs-number">1</span>, cols=<span class="hljs-number">3</span>)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="controlnet" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#controlnet"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->ControlNet<!----></span></h3><!--]--><!----> <p>Mô hình ControlNet được sử dụng cùng với các mô hình khuếch tán khác như Stable Diffusion, và chúng cung cấp một cách linh hoạt và chính xác hơn để kiểm soát cách một hình ảnh được tạo ra. Một ControlNet nhận vào một đầu vào hình ảnh điều kiện bổ sung, giúp định hướng mô hình khuếch tán để giữ lại các đặc điểm có trong hình ảnh đó.</p> <p>Ví dụ, hãy điều kiện hóa một hình ảnh bằng ControlNet được huấn luyện trước trên các hình ảnh inpaint:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np | |
| <span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> ControlNetModel, StableDiffusionControlNetInpaintPipeline | |
| <span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image, make_image_grid | |
| <span class="hljs-comment"># load ControlNet</span> | |
| controlnet = ControlNetModel.from_pretrained(<span class="hljs-string">"lllyasviel/control_v11p_sd15_inpaint"</span>, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span>) | |
| <span class="hljs-comment"># pass ControlNet to the pipeline</span> | |
| pipeline = StableDiffusionControlNetInpaintPipeline.from_pretrained( | |
| <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-inpainting"</span>, controlnet=controlnet, dtype=torch.float16, variant=<span class="hljs-string">"fp16"</span> | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| <span class="hljs-comment"># load base and mask image</span> | |
| init_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"</span>) | |
| mask_image = load_image(<span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png"</span>) | |
| <span class="hljs-comment"># prepare control image</span> | |
| <span class="hljs-keyword">def</span> <span class="hljs-title function_">make_inpaint_condition</span>(<span class="hljs-params">init_image, mask_image</span>): | |
| init_image = np.array(init_image.convert(<span class="hljs-string">"RGB"</span>)).astype(np.float32) / <span class="hljs-number">255.0</span> | |
| mask_image = np.array(mask_image.convert(<span class="hljs-string">"L"</span>)).astype(np.float32) / <span class="hljs-number">255.0</span> | |
| <span class="hljs-keyword">assert</span> init_image.shape[<span class="hljs-number">0</span>:<span class="hljs-number">1</span>] == mask_image.shape[<span class="hljs-number">0</span>:<span class="hljs-number">1</span>], <span class="hljs-string">"image and image_mask must have the same image size"</span> | |
| init_image[mask_image > <span class="hljs-number">0.5</span>] = -<span class="hljs-number">1.0</span> <span class="hljs-comment"># set as masked pixel</span> | |
| init_image = np.expand_dims(init_image, <span class="hljs-number">0</span>).transpose(<span class="hljs-number">0</span>, <span class="hljs-number">3</span>, <span class="hljs-number">1</span>, <span class="hljs-number">2</span>) | |
| init_image = torch.from_numpy(init_image) | |
| <span class="hljs-keyword">return</span> init_image | |
| control_image = make_inpaint_condition(init_image, mask_image)<!----></pre></div><!----> <p>Bây giờ hãy tạo một hình ảnh từ các hình ảnh base, mask và control. Bạn sẽ nhận thấy các đặc điểm của hình ảnh base được giữ lại mạnh mẽ trong hình ảnh được tạo ra.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->prompt = <span class="hljs-string">"concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"</span> | |
| image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, control_image=control_image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, PIL.Image.fromarray(np.uint8(control_image[<span class="hljs-number">0</span>][<span class="hljs-number">0</span>])).convert(<span class="hljs-string">'RGB'</span>), image], rows=<span class="hljs-number">2</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----> <p>Bạn có thể đi xa hơn một bước và kết nối nó với một pipeline image-to-image để áp dụng một <a href="https://huggingface.co/nitrosocke/elden-ring-diffusion" rel="nofollow">phong cách</a> mới:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> AutoPipelineForImage2Image | |
| pipeline = AutoPipelineForImage2Image.from_pretrained( | |
| <span class="hljs-string">"nitrosocke/elden-ring-diffusion"</span>, dtype=torch.float16, | |
| ) | |
| pipeline.enable_model_cpu_offload() | |
| <span class="hljs-comment"># remove following line if xFormers is not installed or you have PyTorch 2.0 or higher installed</span> | |
| pipeline.enable_xformers_memory_efficient_attention() | |
| prompt = <span class="hljs-string">"elden ring style castle"</span> <span class="hljs-comment"># include the token "elden ring style" in the prompt</span> | |
| negative_prompt = <span class="hljs-string">"bad architecture, deformed, disfigured, poor details"</span> | |
| image_elden_ring = pipeline(prompt, negative_prompt=negative_prompt, image=image).images[<span class="hljs-number">0</span>] | |
| make_image_grid([init_image, mask_image, image, image_elden_ring], rows=<span class="hljs-number">2</span>, cols=<span class="hljs-number">2</span>)<!----></pre></div><!----> <div class="flex flex-row gap-4"><div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ảnh ban đầu</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-controlnet.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">ControlNet inpaint</figcaption></div> <div class="flex-1"><img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-img2img.png"/> <figcaption class="mt-2 text-center text-sm text-gray-500">image-to-image</figcaption></div></div> <!--[1--><h2 class="relative group"><a id="optimize" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#optimize"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Tối ưu<!----></span></h2><!--]--><!----> <p>Việc chạy các mô hình diffusion có thể khó khăn và chậm nếu bạn bị giới hạn về tài nguyên, nhưng điều đó không nhất thiết phải xảy ra nếu áp dụng một số mẹo tối ưu hóa. Một trong những tối ưu hóa lớn nhất (và dễ nhất) mà bạn có thể bật là chuyển sang cơ chế attention tiết kiệm bộ nhớ. Nếu bạn đang sử dụng PyTorch 2.0, <a href="../optimization/fp16#scaled-dot-product-attention">scaled-dot product attention</a> sẽ được bật tự động và bạn không cần làm gì thêm. Đối với người dùng không dùng PyTorch 2.0, bạn có thể cài đặt và sử dụng bản triển khai attention tiết kiệm bộ nhớ của <a href="../optimization/attention_backends">xFormers</a>. Cả hai tùy chọn đều giúp giảm mức sử dụng bộ nhớ và tăng tốc suy luận.</p> <p>Bạn cũng có thể offload mô hình sang CPU để tiết kiệm thêm bộ nhớ:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-diff "><!----><span class="hljs-addition">+ pipeline.enable_xformers_memory_efficient_attention()</span> | |
| <span class="hljs-addition">+ pipeline.enable_model_cpu_offload()</span><!----></pre></div><!----> <p>Để tăng tốc mã suy luận của bạn hơn nữa, hãy sử dụng <a href="../optimization/fp16#torchcompile"><code>torch_compile</code></a>. Bạn nên bọc <code>torch.compile</code> quanh thành phần tiêu tốn tài nguyên nhất trong pipeline, thường là UNet:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Đã sao chép</div><!----></button><!----></div> <pre class="language-py "><!---->pipeline.unet = torch.<span class="hljs-built_in">compile</span>(pipeline.unet, mode=<span class="hljs-string">"reduce-overhead"</span>, fullgraph=<span class="hljs-literal">True</span>)<!----></pre></div><!----> <p>Tìm hiểu thêm trong <a href="../optimization/overview#optimization-techniques">Tối ưu hóa và mở rộng</a>.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/inpaint.md" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]--> | |
| <script> | |
| { | |
| __sveltekit_kft2r9 = { | |
| base: "/docs/diffusers/main/en", | |
| assets: "/docs/diffusers/main/en" | |
| }; | |
| const element = document.currentScript.parentElement; | |
| Promise.all([ | |
| import("/docs/diffusers/main/en/_app/immutable/entry/start.B6roO9Zr.js"), | |
| import("/docs/diffusers/main/en/_app/immutable/entry/app.CxxEVG8A.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 15], | |
| data: [null,null], | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 131 kB
- Xet hash:
- 5bbbcfdb2617e045329a39815db282308128c9df1a3cab5c746e2fa5c961f094
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.