Hugging Face's logo

codeShare
/
hidream-01-Image-SDNQ-4bit

Safetensors
Model card Files
xet
 Community
codeShare commited on
Commit
c6494d2
·
verified ·
1 Parent(s): 9449dff

Delete README.md

Browse files
Files changed (1) hide show
  1. README.md +0 -348
README.md DELETED
@@ -1,348 +0,0 @@
1
- ---
2
- license: mit
3
- pipeline_tag: image-text-to-image
4
- library_name: transformers
5
- ---
6
-
7
- # HiDream-O1-Image
8
-
9
- HiDream-O1-Image is a natively unified image generative foundation model built on a Pixel-level Unified Transformer (UiT) without external VAEs or disjoint text encoders, which natively encodes raw pixels, text, and task-specific conditions in a single shared token space — supporting text-to-image, image editing, and subject-driven personalization at up to 2,048 × 2,048.
10
-
11
- ## Project Updates
12
- - 🛠️ **May 13, 2026:** Inference & pipeline updates — accelerated IP inference; the IP pipeline now supports **layout** and **skeleton** conditioning; updated the Dev editing scheduler. For editing tasks we recommend using the **full** model. PyTorch 2.9.x is not recommended due to the [issue](https://github.com/QwenLM/Qwen3-VL/issues/1811)
13
- - 🤗 **May 10, 2026:** Try **HiDream-O1-Image** online on Hugging Face Spaces — [🤗 HiDream-O1-Image](https://huggingface.co/spaces/HiDream-ai/HiDream-O1-Image) and [🤗 HiDream-O1-Image-Dev](https://huggingface.co/spaces/HiDream-ai/HiDream-O1-Image-Dev).
14
- - 📕 **May 10, 2026:** Our **technical report** is now available — [📑 HiDream-O1-Image.pdf](https://github.com/HiDream-ai/HiDream-O1-Image/blob/main/assets/HiDream-O1-Image.pdf).
15
- - 🚀 **May 8, 2026:** We've open-sourced **HiDream-O1-Image (8B)**, including both the undistilled and distilled Dev variants, together with the Reasoning-Driven Prompt Agent.
16
-
17
- > **HiDream-O1-Image (codename: Peanut) debuts at #8 in the Artificial Analysis Text to Image Arena, which is positioned to be the new leading open weights Text to Image model (2026-5-5).**
18
-
19
- <p align="center">
20
- <img src="assets/leaderboard.png" alt="Artificial Analysis Text to Image Arena" width="100%"/>
21
- <br><sub><b>Artificial Analysis Text to Image Arena</b> at up to 2,048 × 2,048.</sub>
22
- </p>
23
-
24
- <p align="center">
25
- <img src="assets/general.webp" alt="General text-to-image generation" width="100%"/>
26
- <br><sub><b>General text-to-image generation</b> at up to 2,048 × 2,048.</sub>
27
- </p>
28
-
29
- <p align="center">
30
- <img src="assets/text-layout.webp" alt="Long-text rendering and layout" width="100%"/>
31
- <br><sub><b>Long-text rendering & layout control</b> — accurate, multi-region, multilingual text.</sub>
32
- </p>
33
-
34
- <p align="center">
35
- <img src="assets/IP_2.jpg" alt="Subject-driven personalization" width="100%"/>
36
- <br><sub><b>Subject-driven personalization</b> — preserve identity / IP across new scenes.</sub>
37
- </p>
38
-
39
- ## Key Features
40
-
41
- - 🧬 **Pixel-Level Unified Transformer** — One end-to-end model on raw pixels, no VAE, no disjoint text encoder.
42
- - 🎨 **One Model, Many Tasks** — Text-to-image, long-text rendering, instruction editing, subject-driven personalization, and storyboard generation in a single architecture.
43
- - 🧠 **Reasoning-Driven Prompt Agent** — Built-in "thinking" agent that resolves implicit knowledge, layout, and text rendering before generation.
44
- - 🖼️ **Native High Resolution** — Direct synthesis up to 2,048 × 2,048 with sharp fine-grained detail.
45
- - ⚡ **Exceptional Efficiency and Versatility at 8B Scale** — With only 8B parameters, achieves performance parity with or even surpasses larger open-source DiTs and leading closed-source models.
46
-
47
- ## Models
48
-
49
- | Name | Script | Inference Steps | HuggingFace Repo |
50
- | :--- | :--- | :---: | :--- |
51
- | HiDream-O1-Image | `inference.py` | 50 | [🤗 HiDream-O1-Image](https://huggingface.co/HiDream-ai/HiDream-O1-Image) |
52
- | HiDream-O1-Image-Dev | `inference.py` | 28 | [🤗 HiDream-O1-Image-Dev](https://huggingface.co/HiDream-ai/HiDream-O1-Image-Dev) |
53
- | Prompt Agent | `prompt_agent.py` | — | [🤗 google/gemma-4-31B-it](https://huggingface.co/google/gemma-4-31B-it) |
54
- | Web Demo | `app.py` | — | — |
55
-
56
- ## Evaluation
57
-
58
- We benchmark HiDream-O1-Image against state-of-the-art open-source and proprietary models on five widely-used evaluation suites covering compositional generation, dense prompt alignment, human preference, complex visual text generation, and long-text rendering. In each table, the **best** result is highlighted in bold and the <u>second-best</u> is underlined. Click any benchmark below to expand or collapse.
59
-
60
- <details open>
61
- <summary><b>GenEval</b> — compositional generation</summary>
62
-
63
- | Model | #Params | Single-Obj | Two-Obj | Count | Color | Position | Attr | Overall |
64
- | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
65
- | Nano Banana 2.0 | – | 1.00 | 0.96 | 0.71 | 0.84 | 0.86 | 0.65 | 0.83 |
66
- | Seedream-4.0 | – | 1.00 | 0.92 | 0.71 | 0.93 | 0.78 | 0.68 | 0.84 |
67
- | GPT Image 1 [High] | – | 0.99 | 0.92 | 0.85 | 0.92 | 0.75 | 0.61 | 0.84 |
68
- | GPT Image 2 | – | 0.99 | 0.98 | 0.85 | 0.93 | 0.85 | 0.77 | 0.89 |
69
- | PixArt | 4.3B + 0.6B | 0.98 | 0.50 | 0.44 | 0.80 | 0.08 | 0.07 | 0.48 |
70
- | Show-o | 1.3B | 0.95 | 0.52 | 0.49 | 0.82 | 0.11 | 0.28 | 0.53 |
71
- | Emu3-Gen | 8B | 0.98 | 0.71 | 0.34 | 0.81 | 0.17 | 0.21 | 0.54 |
72
- | SD3-Medium | 5.5B + 2B | 0.98 | 0.74 | 0.63 | 0.67 | 0.34 | 0.36 | 0.62 |
73
- | JanusFlow | 1.3B | 0.97 | 0.59 | 0.45 | 0.83 | 0.53 | 0.42 | 0.63 |
74
- | FLUX.1 [Dev] | 4.8B + 12B | 0.98 | 0.81 | 0.74 | 0.79 | 0.22 | 0.45 | 0.66 |
75
- | SD3.5 Large | 5.5B + 8.1B | 0.98 | 0.89 | 0.73 | 0.83 | 0.34 | 0.47 | 0.71 |
76
- | Janus-Pro-7B | 7B | 0.99 | 0.89 | 0.59 | 0.90 | 0.79 | 0.66 | 0.80 |
77
- | Z-Image-Turbo | 4B + 6B | 1.00 | 0.95 | 0.77 | 0.89 | 0.65 | 0.68 | 0.82 |
78
- | FLUX.2 [Dev] | 24B + 32B | 1.00 | 0.99 | 0.79 | 0.93 | 0.73 | 0.78 | 0.87 |
79
- | Qwen-Image | 7B + 20B | 0.99 | 0.92 | 0.89 | 0.88 | 0.76 | 0.77 | 0.87 |
80
- | **HiDream-O1-Image** | 8B | 1.00 | 0.99 | 0.79 | 0.89 | 0.93 | 0.78 | <u>0.90</u> |
81
- | **HiDream-O1-Image-Pro** | 200B+ | 1.00 | 0.99 | 0.85 | 0.94 | 0.94 | 0.79 | **0.92** |
82
-
83
- </details>
84
-
85
- <details open>
86
- <summary><b>DPG-Bench</b> — dense prompt alignment</summary>
87
-
88
- | Model | #Params | Global | Entity | Attribute | Relation | Other | Overall |
89
- | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
90
- | GPT Image 1 [High] | – | 88.89 | 88.94 | 89.84 | 92.63 | 90.96 | 85.15 |
91
- | GPT Image 2 | – | 87.27 | 91.91 | 90.85 | 91.59 | 91.58 | 85.98 |
92
- | Nano Banana 2.0 | – | 85.17 | 92.55 | 91.16 | 90.45 | 91.08 | 86.90 |
93
- | Seedream-4.0 | – | 87.17 | 92.41 | 92.29 | 93.33 | 95.48 | 88.63 |
94
- | SD v1.5 | 0.12B + 0.86B | 74.63 | 74.23 | 75.39 | 73.49 | 67.81 | 63.18 |
95
- | PixArt | 4.3B + 0.6B | 74.97 | 79.32 | 78.60 | 82.57 | 76.96 | 71.11 |
96
- | Lumina-Next | 2B + 2B | 82.82 | 88.65 | 86.44 | 80.53 | 81.82 | 74.63 |
97
- | SDXL | 0.81B + 2.6B | 83.27 | 82.43 | 80.91 | 86.76 | 80.41 | 74.65 |
98
- | Hunyuan-DiT | 4.8B + 1.5B | 84.59 | 80.59 | 88.01 | 74.36 | 86.41 | 78.87 |
99
- | Emu3-Gen | 8B | 85.21 | 86.68 | 86.84 | 90.22 | 83.15 | 80.60 |
100
- | DALL-E 3 | – | 90.97 | 89.61 | 88.39 | 90.58 | 89.83 | 83.50 |
101
- | FLUX.1 [Dev] | 4.8B + 12B | 74.35 | 90.00 | 88.96 | 90.87 | 88.33 | 83.84 |
102
- | SD3 Medium | 5.5B + 2B | 87.90 | 91.01 | 88.83 | 80.70 | 88.68 | 84.08 |
103
- | Janus-Pro-7B | 7B | 86.90 | 88.90 | 89.40 | 89.32 | 89.48 | 84.19 |
104
- | Z-Image-Turbo | 4B + 6B | 91.29 | 89.59 | 90.14 | 92.16 | 88.68 | 84.86 |
105
- | HiDream-I1-Full | 13.5B + 17B | 76.44 | 90.22 | 89.48 | 93.74 | 91.83 | 85.89 |
106
- | FLUX.2 [Dev] | 24B + 32B | 92.20 | 91.36 | 93.28 | 93.52 | 89.72 | 87.57 |
107
- | Qwen-Image | 7B + 20B | 91.32 | 91.56 | 92.02 | 94.31 | 92.73 | 88.32 |
108
- | **HiDream-O1-Image** | 8B | 95.15 | 92.32 | 93.74 | 92.88 | 90.25 | <u>89.83</u> |
109
- | **HiDream-O1-Image-Pro** | 200B+ | 94.97 | 95.42 | 92.59 | 90.82 | 89.50 | **90.30** |
110
-
111
- </details>
112
-
113
- <details open>
114
- <summary><b>HPSv3</b> — human preference across 12 categories</summary>
115
-
116
- | Model | #Params | All | Characters | Arts | Design | Architecture | Animals | Natural Scenery | Transportation | Products | Plants | Food | Science | Others |
117
- | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
118
- | Seedream-4.0 | – | 9.32 | 9.83 | 9.20 | 8.83 | 9.95 | 8.99 | 9.40 | 9.58 | 9.12 | 9.26 | 9.75 | 9.11 | 9.51 |
119
- | Nano Banana 2.0 | – | 10.01 | 10.18 | 9.18 | 9.58 | 10.96 | 9.71 | 10.04 | 10.38 | 10.36 | 10.14 | 10.61 | 9.14 | 9.89 |
120
- | GPT Image 2 | – | 10.21 | 10.75 | 9.91 | 10.15 | 10.59 | 10.05 | 10.29 | 10.17 | 10.26 | 10.07 | 10.75 | 10.05 | 10.00 |
121
- | Z-Image-Turbo | 4B + 6B | 8.35 | 8.98 | 8.29 | 7.65 | 9.26 | 8.51 | 8.33 | 8.81 | 7.83 | 8.46 | 8.64 | 7.93 | 8.57 |
122
- | FLUX.2 [Dev] | 24B + 32B | 9.28 | 10.23 | 9.56 | 8.80 | 9.73 | 9.43 | 9.21 | 9.44 | 8.93 | 9.23 | 9.82 | 8.67 | 9.11 |
123
- | Qwen-Image | 7B + 20B | 9.94 | 10.91 | 10.47 | 9.56 | 10.22 | 10.61 | 9.87 | 10.10 | 9.15 | 9.99 | 10.08 | 9.19 | 9.83 |
124
- | **HiDream-O1-Image** | 8B | <u>10.37</u> | 10.59 | 10.44 | 10.29 | 11.02 | 10.34 | 10.37 | 10.54 | 10.50 | 10.38 | 10.85 | 9.68 | 10.09 |
125
- | **HiDream-O1-Image-Pro** | 200B+ | **10.47** | 10.63 | 10.51 | 10.33 | 11.11 | 10.08 | 10.45 | 10.37 | 10.75 | 10.29 | 11.13 | 10.09 | 10.39 |
126
-
127
- </details>
128
-
129
- <details>
130
- <summary><b>CVTG-2K</b> — complex visual text generation (click to expand)</summary>
131
-
132
- | Model | #Params | 2 regions | 3 regions | 4 regions | 5 regions | Average | NED | CLIP Score |
133
- | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
134
- | Nano Banana 2.0 | – | 0.7465 | 0.7720 | 0.8067 | 0.7980 | 0.7875 | 0.8945 | 0.7212 |
135
- | GPT Image 1 [High] | – | 0.8779 | 0.8659 | 0.8731 | 0.8218 | 0.8569 | 0.9478 | 0.7982 |
136
- | Seedream-4.0 | – | 0.8980 | 0.8949 | 0.9044 | 0.9015 | 0.9003 | 0.9511 | 0.8033 |
137
- | GPT Image 2 | – | 0.8904 | 0.8887 | 0.9101 | 0.9044 | 0.9003 | 0.9515 | 0.7798 |
138
- | TextDiffuser-2 | 0.12B + 0.9B | 0.5322 | 0.3255 | 0.1787 | 0.0809 | 0.2326 | 0.4353 | 0.6765 |
139
- | RAG-Diffusion | 4.8B + 12B | 0.4388 | 0.3316 | 0.2116 | 0.1910 | 0.2648 | 0.4498 | 0.7797 |
140
- | AnyText | 0.123B + 1.2B | 0.0513 | 0.1739 | 0.1948 | 0.2249 | 0.1804 | 0.4675 | 0.7432 |
141
- | 3DIS | 0.81B + 2.6B | 0.4495 | 0.3959 | 0.3880 | 0.3303 | 0.3813 | 0.6505 | 0.7767 |
142
- | FLUX.1 [Dev] | 4.8B + 12B | 0.6089 | 0.5531 | 0.4661 | 0.4316 | 0.4965 | 0.6879 | 0.7401 |
143
- | SD3.5 Large | 5.5B + 8.1B | 0.7293 | 0.6825 | 0.6574 | 0.5940 | 0.6548 | 0.8470 | 0.7797 |
144
- | TextCrafter | 7B + 20B | 0.7628 | 0.7628 | 0.7406 | 0.6977 | 0.7370 | 0.8679 | 0.7868 |
145
- | Qwen-Image | 7B + 20B | 0.8370 | 0.8364 | 0.8313 | 0.8158 | 0.8288 | 0.9116 | 0.8017 |
146
- | Z-Image-Turbo | 4B + 6B | 0.8872 | 0.8662 | 0.8628 | 0.8347 | 0.8585 | 0.9281 | 0.8048 |
147
- | FLUX.2 [Dev] | 24B + 32B | 0.9261 | 0.8897 | 0.8995 | 0.8732 | 0.8926 | 0.9475 | <u>0.8104</u> |
148
- | **HiDream-O1-Image** | 8B | 0.9085 | 0.9159 | 0.9216 | 0.9015 | <u>0.9128</u> | <u>0.9561</u> | 0.8076 |
149
- | **HiDream-O1-Image-Pro** | 200B+ | 0.9133 | 0.9221 | 0.9365 | 0.9175 | **0.9222** | **0.9628** | **0.8349** |
150
-
151
- </details>
152
-
153
- <details>
154
- <summary><b>LongText-Bench</b> — long-text rendering, EN & ZH (click to expand)</summary>
155
-
156
- | Model | #Params | LongText-Bench-EN | LongText-Bench-ZH |
157
- | :--- | :---: | :---: | :---: |
158
- | Seedream-4.0 | – | 0.936 | 0.946 |
159
- | GPT Image 1 [High] | – | 0.956 | 0.619 |
160
- | GPT Image 2 | – | 0.960 | 0.961 |
161
- | Nano Banana 2.0 | – | <u>0.980</u> | 0.965 |
162
- | Janus-Pro-7B | 7B | 0.019 | 0.006 |
163
- | BLIP3-o | 7B + 1.4B | 0.021 | 0.018 |
164
- | Kolors 2.0 | – | 0.258 | 0.329 |
165
- | BAGEL | 7B + 7B | 0.373 | 0.310 |
166
- | OmniGen2 | 3B + 4B | 0.561 | 0.059 |
167
- | X-Omni | 7B | 0.900 | 0.814 |
168
- | HiDream-I1-Full | 13.5B + 17B | 0.543 | 0.024 |
169
- | FLUX.1 [Dev] | 4.8B + 12B | 0.607 | 0.005 |
170
- | Z-Image-Turbo | 4B + 6B | 0.917 | 0.926 |
171
- | FLUX.2 [Dev] | 24B + 32B | 0.963 | 0.757 |
172
- | Qwen-Image | 7B + 20B | 0.943 | 0.946 |
173
- | **HiDream-O1-Image** | 8B | 0.979 | <u>0.978</u> |
174
- | **HiDream-O1-Image-Pro** | 200B+ | **0.982** | **0.980** |
175
-
176
- </details>
177
-
178
- ## Installation
179
-
180
- 1. Clone this repository:
181
- ```bash
182
- git clone https://github.com/HiDream-ai/HiDream-O1-Image.git
183
- cd HiDream-O1-Image
184
- ```
185
-
186
- 2. Install the required dependencies:
187
- ```bash
188
- pip install -r requirements.txt
189
- ```
190
-
191
- > **Note on `flash-attn`.** We highly recommend installing [`flash-attn`](https://github.com/Dao-AILab/flash-attention) for optimized attention computation. **If you do not (or cannot) install `flash-attn`, you must edit `models/pipeline.py` line 291 and change `"use_flash_attn": True` to `"use_flash_attn": False`** — otherwise inference will fail to import the kernel.
192
-
193
- ## Reasoning-Driven Prompt Agent
194
-
195
- HiDream-O1-Image ships with a Reasoning-Driven Prompt Agent (`prompt_agent.py`) that explicitly reasons through layout, subject attributes, physical logic, and text-rendering details, then rewrites a raw user instruction into a self-contained English prompt. It supports two backends — pick one with `--backend`.
196
-
197
- The agent prints a JSON object with three fields: `prompt` (rewritten English prompt), `reasoning`, and `resolved_knowledge`. Feed the `prompt` field into `inference.py` for best results on intricate, reasoning-heavy requests.
198
-
199
- ### Option A — Local Backend (Gemma-4-31B-it)
200
-
201
- 1. Download the Gemma weights (requires accepting the Gemma license on HuggingFace):
202
- ```bash
203
- huggingface-cli download google/gemma-4-31B-it --local-dir /path/to/gemma-4-31B-it
204
- ```
205
-
206
- 2. Run the refiner locally:
207
- ```bash
208
- python prompt_agent.py \
209
- --backend local \
210
- --model_id /path/to/gemma-4-31B-it \
211
- --prompt "李白的静夜思写在古墙上"
212
- ```
213
-
214
- ### Option B — External OpenAI-Compatible API
215
-
216
- Use any OpenAI-compatible endpoint (OpenAI, Azure, vLLM, SGLang, DeepSeek, etc.) by providing `--base_url`, `--api_key`, and `--model_name`:
217
-
218
- ```bash
219
- python prompt_agent.py \
220
- --backend api \
221
- --base_url https://api.openai.com/v1 \
222
- --api_key $OPENAI_API_KEY \
223
- --model_name deepseek-v4-pro \
224
- --prompt "李白的静夜思写在古墙上"
225
- ```
226
-
227
- ## Usage
228
-
229
- A CUDA-capable GPU is required for inference. The examples below use the **undistilled** model (`--model_type full`); see the last subsection for running the same tasks with the **distilled** model (`--model_type dev`).
230
-
231
- ### 1. Text-to-Image Generation
232
- Generate an image from a text prompt:
233
-
234
- ```bash
235
- python inference.py \
236
- --model_path /path/to/HiDream-O1-Image \
237
- --prompt "medium shot, eye-level, front view. A woman is seated in an ornate bedroom, illuminated by candlelight, with a calm and composed expression. The subject is a young woman with fair skin, light brown hair styled in an updo with loose tendrils framing her face, and blue eyes. She wears a cream-colored satin robe with delicate floral embroidery and lace trim along the neckline. Her ears are adorned with pearl drop earrings. She is seated on a bed with a dark, intricately carved wooden headboard. To her left, a wooden nightstand holds three lit white candles and a candelabra with multiple lit candles in the background. The bed is covered with patterned pillows and a dark, textured blanket. The walls are paneled with dark wood and feature a large, ornate tapestry with muted earth tones. The lighting creates soft highlights on her face and robe, with warm shadows cast across the room." \
238
- --output_image results/t2i.png \
239
- --height 2048 \
240
- --width 2048
241
- ```
242
-
243
- ### 2. Instruction-Based Image Editing
244
- Provide a single reference image and an editing instruction:
245
-
246
- ```bash
247
- python inference.py \
248
- --model_path /path/to/HiDream-O1-Image \
249
- --prompt "remove the earphones" \
250
- --ref_images assets/edit/test.jpg \
251
- --output_image results/edit.png \
252
- --keep_original_aspect
253
- ```
254
-
255
- ### 3. Multi-Reference Subject-Driven Personalization
256
- Provide two or more reference images that define the subject(s), and a prompt that places them in a new scene:
257
-
258
- ```bash
259
- python inference.py \
260
- --model_path /path/to/HiDream-O1-Image \
261
- --shift 1 \
262
- --prompt "A young boy with blonde hair stands on steps wearing light blue jeans, a white t-shirt with logo, and blue and white sneakers. He wears a brown cord necklace with beads, a black wristwatch with digital display, and carries a yellow fanny pack with white zipper. In his hand is a red boxing glove with white top, a teal plastic toy car, and a plastic toy figure of Captain America. He wears a straw hat with cream band. Natural light illuminates the scene." \
263
- --ref_images assets/IP/1.jpg assets/IP/2.jpg assets/IP/3.jpg assets/IP/4.jpg assets/IP/5.jpg assets/IP/6.jpg assets/IP/7.jpg assets/IP/8.jpg assets/IP/9.jpg assets/IP/10.jpg \
264
- --output_image results/subject.png
265
- ```
266
-
267
- ### 4. Multi-Reference Subject-Driven Personalization with Skeleton
268
- ```bash
269
- python inference.py \
270
- --model_path /path/to/HiDream-O1-Image \
271
- --shift 1 \
272
- --seed 42 \
273
- --prompt "Create a realistic try-on image of the person wearing the provided clothing." \
274
- --ref_images assets/IP_skeleton/0.face.jpg assets/IP_skeleton/0.bg.jpg assets/IP_skeleton/0.openpose.jpg assets/IP_skeleton/0.part_1.jpg assets/IP_skeleton/0.part_2.jpg assets/IP_skeleton/0.part_3.jpg \
275
- --output_image results/subject.png
276
- ```
277
-
278
- ### 5. Multi-Reference Subject-Driven Personalization with Layout
279
- ```bash
280
- python inference.py \
281
- --model_path /path/to/HiDream-O1-Image \
282
- --shift 1 \
283
- --seed 42 \
284
- --prompt "City council members pose with relaxed smiles on a sunlit terrace, warm approachable mood, golden hour, cinematic soft glow." \
285
- --ref_images assets/IP_layout/0.jpg assets/IP_layout/1.jpg \
286
- --layout_bboxes "[[0.20507812, 0.43945312, 0.48828125, 0.7421875 ], [0.57617188, 0.80078125, 0.08789062, 0.34179688]]" \
287
- --output_image results/ip_layout.png
288
- ```
289
-
290
- ### 6. Running with the Dev Model
291
- All three tasks above can be run with the **Dev** model by switching `--model_path` to the Dev checkpoint and setting `--model_type dev`. For example:
292
-
293
- ```bash
294
- python inference.py \
295
- --model_path /path/to/HiDream-O1-Image-Dev \
296
- --prompt "A dog holds a sign that says \"HiDream-O1-Image release.\"" \
297
- --output_image results/t2i_dev.png \
298
- --model_type dev
299
- ```
300
-
301
- ### Command Line Arguments
302
-
303
- - `--model_path`: Path to the complete HuggingFace model directory (undistilled or distilled).
304
- - `--prompt`: Text prompt for the generation or editing task.
305
- - `--ref_images`: Paths to one or more reference images (optional; space-separated).
306
- - `--output_image`: Path to save the generated image (default: `output.png`).
307
- - `--height` / `--width`: Output image dimensions (default: `2048` × `2048`; values snap to valid resolutions internally).
308
- - `--model_type`: `full` or `dev` (default: `full`). Selects the inference recipe:
309
- - `full`: 50 steps, guidance scale `5.0`, shift `3.0`, default scheduler.
310
- - `dev`: 28 steps, guidance scale `0.0`, shift `1.0`, flash scheduler with predefined timesteps.
311
- - `--seed`: Random seed (default: `32`).
312
- - `--guidance_scale`: Guidance scale (default: `5.0`). Only effective when `--model_type` is `full`.
313
- - `--noise_scale_start`, `--noise_scale_end`: Control the scale of the noise injected by the scheduler at each denoising step; the per-step scale linearly interpolates from `noise_scale_start` (first step) to `noise_scale_end` (last step). See `models/pipeline.py:262` and `models/pipeline.py:273`. Defaults: `7.5`, `7.5`.
314
- - `--noise_clip_std`: Per-step clipping threshold (in units of the injected noise's standard deviation) applied to the noise added during scheduler stepping. See `models/flash_scheduler.py:348-350`. Default: `2.5`.
315
- - `--keep_original_aspect`: When exactly one reference image is provided, resize it with `max_size=2048` and use its dimensions for the target image (preserves the reference's aspect ratio) if `True`.
316
-
317
- ## Web Demo
318
-
319
- `app.py` is a self-contained Flask web application that exposes all generation modes. It also integrates the Reasoning-Driven Prompt Agent.
320
-
321
- ### Starting the server
322
-
323
- ```bash
324
- python app.py \
325
- --model_path /path/to/HiDream-O1-Image \
326
- --host 0.0.0.0 \
327
- --port 7860
328
- ```
329
-
330
- Then open `http://localhost:7860` in your browser.
331
-
332
- ### Command-line arguments
333
-
334
- | Argument | Default | Description |
335
- | :--- | :--- | :--- |
336
- | `--model_path` | `$HIDREAM_MODEL_PATH` | Path to the checkpoint directory (`HiDream-O1-Image` or `HiDream-O1-Image-Dev`). |
337
- | `--model_type` | `full` | `full` (50-step) or `dev` (28-step). |
338
- | `--host` | `0.0.0.0` | Bind address for the Flask server. |
339
- | `--port` | `7860` | Port for the Flask server. |
340
-
341
- All four arguments can also be set via environment variables (see `.env.example`): `HIDREAM_MODEL_PATH`, `HIDREAM_MODEL_TYPE`, `HIDREAM_HOST`, and `HIDREAM_PORT`.
342
-
343
- ### Prompt Agent in the UI
344
-
345
- The sidebar contains a Prompt Agent panel that calls the same Reasoning-Driven Prompt Agent used by `prompt_agent.py`. Select either the *OpenAI-compatible API* backend (any endpoint, key, and model name) or the *Local · Gemma* backend (set `HIDREAM_AGENT_MODEL` in `.env` or the environment to point to your local Gemma-4-31B-it weights).
346
-
347
- ## License
348
- The code in this repository and the HiDream-O1-Image models are licensed under MIT License.