--- license: mit base_model: runwayml/stable-diffusion-v1-5 library_name: diffusers pipeline_tag: text-to-image tags: - stable-diffusion - lora - diffusers - style-transfer - cubism - post-impressionism - ukiyo-e --- # SD 1.5 Multi-Style LoRA Pack Lightweight rank-4 UNet LoRA adapters (`3.08 MiB` each) trained on Stable Diffusion 1.5 for artistic style transfer, powering [darrige.tech](https://darrige.tech). ## Included Styles & Triggers | Style | Subfolder | Trigger Phrase | | :--- | :--- | :--- | | **Cubism** | `cubism` | `in cubism style, geometric shapes, picasso style` | | **Post-Impressionism** | `post-impressionism` | `in post-impressionism style, painting style` | | **Ukiyo-e** | `ukiyo-e` | `in ukiyo-e style, woodblock print, flat colors` | ## Quickstart (`diffusers`) ```python import torch from diffusers import StableDiffusionPipeline, AutoencoderKL vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16) pipe = StableDiffusionPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", vae=vae, torch_dtype=torch.float16, ).to("cuda") # Choose subfolder: "cubism", "post-impressionism", or "ukiyo-e" pipe.load_lora_weights("Darrige/sd15-multistyle-lora", subfolder="ukiyo-e") pipe.fuse_lora(lora_scale=1.05) image = pipe( prompt="a quiet mountain temple at dawn, in ukiyo-e style, woodblock print, flat colors", negative_prompt="photograph, photorealistic, blurry, watermark, 3d render, cg", num_inference_steps=30, guidance_scale=7.5, ).images[0] image.save("styled.png") ``` **Recommended Img2Img Settings:** `EulerAncestralDiscreteScheduler`, `strength` `0.42 - 0.62`, `lora_scale` `0.95 - 1.30`, `guidance_scale` `8.0`, `50` steps. ## Warnings & Limitations - **SD 1.5 Only:** Incompatible with SDXL, SD3, or Flux. - **Adapters Only:** Requires downloading `runwayml/stable-diffusion-v1-5` at runtime. - **Safety:** These are raw LoRA weights without a built-in safety filter; enable a safety checker in production deployments. ## Links - **Live Portfolio:** [darrige.tech](https://darrige.tech) - **Source Code:** [GitHub](https://github.com/Mohammed-Darrige/neural-style-transfer) - **Author:** [Mohammed Darrige](https://www.linkedin.com/in/mohammed-darrige/)