[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M73JD7P968NE0DFN1VXK070S":3,"skill-same:01M73JD7P968NE0DFN1VXK070S":80},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":38,"commitSha":39,"skillDirectory":40,"forksCount":41,"repositoryHtmlUrl":42,"repositoryHomepageUrl":6,"ownerHtmlUrl":43,"ownerBlogUrl":44,"ownerDescription":6,"updatedAt":45,"install":46,"resources":59},"01M73JD7P968NE0DFN1VXK070S","stable-diffusion-image-generation",null,"State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.","qhjqhj00\u002Fresearch-skills-pool","qhjqhj00","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F29684092?v=4","01M72AE0ZGXVM5C57N512DGQG8","Tommy Chien","01M72AE107ESJMHT7QC5SXX144",3,"MIT","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: stable-diffusion-image-generation\ndescription: State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.\nversion: 1.0.0\nauthor: Orchestra Research\nlicense: MIT\ntags: [Image Generation, Stable Diffusion, Diffusers, Text-to-Image, Multimodal, Computer Vision]\ndependencies: [diffusers>=0.30.0, transformers>=4.41.0, accelerate>=0.31.0, torch>=2.0.0]\n---\n\n# Stable Diffusion Image Generation\n\nComprehensive guide to generating images with Stable Diffusion using the HuggingFace Diffusers library.\n\n## When to use Stable Diffusion\n\n**Use Stable Diffusion when:**\n- Generating images from text descriptions\n- Performing image-to-image translation (style transfer, enhancement)\n- Inpainting (filling in masked regions)\n- Outpainting (extending images beyond boundaries)\n- Creating variations of existing images\n- Building custom image generation workflows\n\n**Key features:**\n- **Text-to-Image**: Generate images from natural language prompts\n- **Image-to-Image**: Transform existing images with text guidance\n- **Inpainting**: Fill masked regions with context-aware content\n- **ControlNet**: Add spatial conditioning (edges, poses, depth)\n- **LoRA Support**: Efficient fine-tuning and style adaptation\n- **Multiple Models**: SD 1.5, SDXL, SD 3.0, Flux support\n\n**Use alternatives instead:**\n- **DALL-E 3**: For API-based generation without GPU\n- **Midjourney**: For artistic, stylized outputs\n- **Imagen**: For Google Cloud integration\n- **Leonardo.ai**: For web-based creative workflows\n\n## Quick start\n\n### Installation\n\n```bash\npip install diffusers transformers accelerate torch\npip install xformers  # Optional: memory-efficient attention\n```\n\n### Basic text-to-image\n\n```python\nfrom diffusers import DiffusionPipeline\nimport torch\n\n# Load pipeline (auto-detects model type)\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n)\npipe.to(\"cuda\")\n\n# Generate image\nimage = pipe(\n    \"A serene mountain landscape at sunset, highly detailed\",\n    num_inference_steps=50,\n    guidance_scale=7.5\n).images[0]\n\nimage.save(\"output.png\")\n```\n\n### Using SDXL (higher quality)\n\n```python\nfrom diffusers import AutoPipelineForText2Image\nimport torch\n\npipe = AutoPipelineForText2Image.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\npipe.to(\"cuda\")\n\n# Enable memory optimization\npipe.enable_model_cpu_offload()\n\nimage = pipe(\n    prompt=\"A futuristic city with flying cars, cinematic lighting\",\n    height=1024,\n    width=1024,\n    num_inference_steps=30\n).images[0]\n```\n\n## Architecture overview\n\n### Three-pillar design\n\nDiffusers is built around three core components:\n\n```\nPipeline (orchestration)\n├── Model (neural networks)\n│   ├── UNet \u002F Transformer (noise prediction)\n│   ├── VAE (latent encoding\u002Fdecoding)\n│   └── Text Encoder (CLIP\u002FT5)\n└── Scheduler (denoising algorithm)\n```\n\n### Pipeline inference flow\n\n```\nText Prompt → Text Encoder → Text Embeddings\n                                    ↓\nRandom Noise → [Denoising Loop] ← Scheduler\n                      ↓\n               Predicted Noise\n                      ↓\n              VAE Decoder → Final Image\n```\n\n## Core concepts\n\n### Pipelines\n\nPipelines orchestrate complete workflows:\n\n| Pipeline | Purpose |\n|----------|---------|\n| `StableDiffusionPipeline` | Text-to-image (SD 1.x\u002F2.x) |\n| `StableDiffusionXLPipeline` | Text-to-image (SDXL) |\n| `StableDiffusion3Pipeline` | Text-to-image (SD 3.0) |\n| `FluxPipeline` | Text-to-image (Flux models) |\n| `StableDiffusionImg2ImgPipeline` | Image-to-image |\n| `StableDiffusionInpaintPipeline` | Inpainting |\n\n### Schedulers\n\nSchedulers control the denoising process:\n\n| Scheduler | Steps | Quality | Use Case |\n|-----------|-------|---------|----------|\n| `EulerDiscreteScheduler` | 20-50 | Good | Default choice |\n| `EulerAncestralDiscreteScheduler` | 20-50 | Good | More variation |\n| `DPMSolverMultistepScheduler` | 15-25 | Excellent | Fast, high quality |\n| `DDIMScheduler` | 50-100 | Good | Deterministic |\n| `LCMScheduler` | 4-8 | Good | Very fast |\n| `UniPCMultistepScheduler` | 15-25 | Excellent | Fast convergence |\n\n### Swapping schedulers\n\n```python\nfrom diffusers import DPMSolverMultistepScheduler\n\n# Swap for faster generation\npipe.scheduler = DPMSolverMultistepScheduler.from_config(\n    pipe.scheduler.config\n)\n\n# Now generate with fewer steps\nimage = pipe(prompt, num_inference_steps=20).images[0]\n```\n\n## Generation parameters\n\n### Key parameters\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `prompt` | Required | Text description of desired image |\n| `negative_prompt` | None | What to avoid in the image |\n| `num_inference_steps` | 50 | Denoising steps (more = better quality) |\n| `guidance_scale` | 7.5 | Prompt adherence (7-12 typical) |\n| `height`, `width` | 512\u002F1024 | Output dimensions (multiples of 8) |\n| `generator` | None | Torch generator for reproducibility |\n| `num_images_per_prompt` | 1 | Batch size |\n\n### Reproducible generation\n\n```python\nimport torch\n\ngenerator = torch.Generator(device=\"cuda\").manual_seed(42)\n\nimage = pipe(\n    prompt=\"A cat wearing a top hat\",\n    generator=generator,\n    num_inference_steps=50\n).images[0]\n```\n\n### Negative prompts\n\n```python\nimage = pipe(\n    prompt=\"Professional photo of a dog in a garden\",\n    negative_prompt=\"blurry, low quality, distorted, ugly, bad anatomy\",\n    guidance_scale=7.5\n).images[0]\n```\n\n## Image-to-image\n\nTransform existing images with text guidance:\n\n```python\nfrom diffusers import AutoPipelineForImage2Image\nfrom PIL import Image\n\npipe = AutoPipelineForImage2Image.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\ninit_image = Image.open(\"input.jpg\").resize((512, 512))\n\nimage = pipe(\n    prompt=\"A watercolor painting of the scene\",\n    image=init_image,\n    strength=0.75,  # How much to transform (0-1)\n    num_inference_steps=50\n).images[0]\n```\n\n## Inpainting\n\nFill masked regions:\n\n```python\nfrom diffusers import AutoPipelineForInpainting\nfrom PIL import Image\n\npipe = AutoPipelineForInpainting.from_pretrained(\n    \"runwayml\u002Fstable-diffusion-inpainting\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\nimage = Image.open(\"photo.jpg\")\nmask = Image.open(\"mask.png\")  # White = inpaint region\n\nresult = pipe(\n    prompt=\"A red car parked on the street\",\n    image=image,\n    mask_image=mask,\n    num_inference_steps=50\n).images[0]\n```\n\n## ControlNet\n\nAdd spatial conditioning for precise control:\n\n```python\nfrom diffusers import StableDiffusionControlNetPipeline, ControlNetModel\nimport torch\n\n# Load ControlNet for edge conditioning\ncontrolnet = ControlNetModel.from_pretrained(\n    \"lllyasviel\u002Fcontrol_v11p_sd15_canny\",\n    torch_dtype=torch.float16\n)\n\npipe = StableDiffusionControlNetPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    controlnet=controlnet,\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Use Canny edge image as control\ncontrol_image = get_canny_image(input_image)\n\nimage = pipe(\n    prompt=\"A beautiful house in the style of Van Gogh\",\n    image=control_image,\n    num_inference_steps=30\n).images[0]\n```\n\n### Available ControlNets\n\n| ControlNet | Input Type | Use Case |\n|------------|------------|----------|\n| `canny` | Edge maps | Preserve structure |\n| `openpose` | Pose skeletons | Human poses |\n| `depth` | Depth maps | 3D-aware generation |\n| `normal` | Normal maps | Surface details |\n| `mlsd` | Line segments | Architectural lines |\n| `scribble` | Rough sketches | Sketch-to-image |\n\n## LoRA adapters\n\nLoad fine-tuned style adapters:\n\n```python\nfrom diffusers import DiffusionPipeline\n\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Load LoRA weights\npipe.load_lora_weights(\"path\u002Fto\u002Flora\", weight_name=\"style.safetensors\")\n\n# Generate with LoRA style\nimage = pipe(\"A portrait in the trained style\").images[0]\n\n# Adjust LoRA strength\npipe.fuse_lora(lora_scale=0.8)\n\n# Unload LoRA\npipe.unload_lora_weights()\n```\n\n### Multiple LoRAs\n\n```python\n# Load multiple LoRAs\npipe.load_lora_weights(\"lora1\", adapter_name=\"style\")\npipe.load_lora_weights(\"lora2\", adapter_name=\"character\")\n\n# Set weights for each\npipe.set_adapters([\"style\", \"character\"], adapter_weights=[0.7, 0.5])\n\nimage = pipe(\"A portrait\").images[0]\n```\n\n## Memory optimization\n\n### Enable CPU offloading\n\n```python\n# Model CPU offload - moves models to CPU when not in use\npipe.enable_model_cpu_offload()\n\n# Sequential CPU offload - more aggressive, slower\npipe.enable_sequential_cpu_offload()\n```\n\n### Attention slicing\n\n```python\n# Reduce memory by computing attention in chunks\npipe.enable_attention_slicing()\n\n# Or specific chunk size\npipe.enable_attention_slicing(\"max\")\n```\n\n### xFormers memory-efficient attention\n\n```python\n# Requires xformers package\npipe.enable_xformers_memory_efficient_attention()\n```\n\n### VAE slicing for large images\n\n```python\n# Decode latents in tiles for large images\npipe.enable_vae_slicing()\npipe.enable_vae_tiling()\n```\n\n## Model variants\n\n### Loading different precisions\n\n```python\n# FP16 (recommended for GPU)\npipe = DiffusionPipeline.from_pretrained(\n    \"model-id\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\n\n# BF16 (better precision, requires Ampere+ GPU)\npipe = DiffusionPipeline.from_pretrained(\n    \"model-id\",\n    torch_dtype=torch.bfloat16\n)\n```\n\n### Loading specific components\n\n```python\nfrom diffusers import UNet2DConditionModel, AutoencoderKL\n\n# Load custom VAE\nvae = AutoencoderKL.from_pretrained(\"stabilityai\u002Fsd-vae-ft-mse\")\n\n# Use with pipeline\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    vae=vae,\n    torch_dtype=torch.float16\n)\n```\n\n## Batch generation\n\nGenerate multiple images efficiently:\n\n```python\n# Multiple prompts\nprompts = [\n    \"A cat playing piano\",\n    \"A dog reading a book\",\n    \"A bird painting a picture\"\n]\n\nimages = pipe(prompts, num_inference_steps=30).images\n\n# Multiple images per prompt\nimages = pipe(\n    \"A beautiful sunset\",\n    num_images_per_prompt=4,\n    num_inference_steps=30\n).images\n```\n\n## Common workflows\n\n### Workflow 1: High-quality generation\n\n```python\nfrom diffusers import StableDiffusionXLPipeline, DPMSolverMultistepScheduler\nimport torch\n\n# 1. Load SDXL with optimizations\npipe = StableDiffusionXLPipeline.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\npipe.to(\"cuda\")\npipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)\npipe.enable_model_cpu_offload()\n\n# 2. Generate with quality settings\nimage = pipe(\n    prompt=\"A majestic lion in the savanna, golden hour lighting, 8k, detailed fur\",\n    negative_prompt=\"blurry, low quality, cartoon, anime, sketch\",\n    num_inference_steps=30,\n    guidance_scale=7.5,\n    height=1024,\n    width=1024\n).images[0]\n```\n\n### Workflow 2: Fast prototyping\n\n```python\nfrom diffusers import AutoPipelineForText2Image, LCMScheduler\nimport torch\n\n# Use LCM for 4-8 step generation\npipe = AutoPipelineForText2Image.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Load LCM LoRA for fast generation\npipe.load_lora_weights(\"latent-consistency\u002Flcm-lora-sdxl\")\npipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)\npipe.fuse_lora()\n\n# Generate in ~1 second\nimage = pipe(\n    \"A beautiful landscape\",\n    num_inference_steps=4,\n    guidance_scale=1.0\n).images[0]\n```\n\n## Common issues\n\n**CUDA out of memory:**\n```python\n# Enable memory optimizations\npipe.enable_model_cpu_offload()\npipe.enable_attention_slicing()\npipe.enable_vae_slicing()\n\n# Or use lower precision\npipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)\n```\n\n**Black\u002Fnoise images:**\n```python\n# Check VAE configuration\n# Use safety checker bypass if needed\npipe.safety_checker = None\n\n# Ensure proper dtype consistency\npipe = pipe.to(dtype=torch.float16)\n```\n\n**Slow generation:**\n```python\n# Use faster scheduler\nfrom diffusers import DPMSolverMultistepScheduler\npipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)\n\n# Reduce steps\nimage = pipe(prompt, num_inference_steps=20).images[0]\n```\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Custom pipelines, fine-tuning, deployment\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fdiffusers\n- **Repository**: https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fdiffusers\n- **Model Hub**: https:\u002F\u002Fhuggingface.co\u002Fmodels?library=diffusers\n- **Discord**: https:\u002F\u002Fdiscord.gg\u002Fdiffusers\n","# Stable Diffusion Image Generation\n\nComprehensive guide to generating images with Stable Diffusion using the HuggingFace Diffusers library.\n\n## When to use Stable Diffusion\n\n**Use Stable Diffusion when:**\n- Generating images from text descriptions\n- Performing image-to-image translation (style transfer, enhancement)\n- Inpainting (filling in masked regions)\n- Outpainting (extending images beyond boundaries)\n- Creating variations of existing images\n- Building custom image generation workflows\n\n**Key features:**\n- **Text-to-Image**: Generate images from natural language prompts\n- **Image-to-Image**: Transform existing images with text guidance\n- **Inpainting**: Fill masked regions with context-aware content\n- **ControlNet**: Add spatial conditioning (edges, poses, depth)\n- **LoRA Support**: Efficient fine-tuning and style adaptation\n- **Multiple Models**: SD 1.5, SDXL, SD 3.0, Flux support\n\n**Use alternatives instead:**\n- **DALL-E 3**: For API-based generation without GPU\n- **Midjourney**: For artistic, stylized outputs\n- **Imagen**: For Google Cloud integration\n- **Leonardo.ai**: For web-based creative workflows\n\n## Quick start\n\n### Installation\n\n```bash\npip install diffusers transformers accelerate torch\npip install xformers  # Optional: memory-efficient attention\n```\n\n### Basic text-to-image\n\n```python\nfrom diffusers import DiffusionPipeline\nimport torch\n\n# Load pipeline (auto-detects model type)\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n)\npipe.to(\"cuda\")\n\n# Generate image\nimage = pipe(\n    \"A serene mountain landscape at sunset, highly detailed\",\n    num_inference_steps=50,\n    guidance_scale=7.5\n).images[0]\n\nimage.save(\"output.png\")\n```\n\n### Using SDXL (higher quality)\n\n```python\nfrom diffusers import AutoPipelineForText2Image\nimport torch\n\npipe = AutoPipelineForText2Image.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\npipe.to(\"cuda\")\n\n# Enable memory optimization\npipe.enable_model_cpu_offload()\n\nimage = pipe(\n    prompt=\"A futuristic city with flying cars, cinematic lighting\",\n    height=1024,\n    width=1024,\n    num_inference_steps=30\n).images[0]\n```\n\n## Architecture overview\n\n### Three-pillar design\n\nDiffusers is built around three core components:\n\n```\nPipeline (orchestration)\n├── Model (neural networks)\n│   ├── UNet \u002F Transformer (noise prediction)\n│   ├── VAE (latent encoding\u002Fdecoding)\n│   └── Text Encoder (CLIP\u002FT5)\n└── Scheduler (denoising algorithm)\n```\n\n### Pipeline inference flow\n\n```\nText Prompt → Text Encoder → Text Embeddings\n                                    ↓\nRandom Noise → [Denoising Loop] ← Scheduler\n                      ↓\n               Predicted Noise\n                      ↓\n              VAE Decoder → Final Image\n```\n\n## Core concepts\n\n### Pipelines\n\nPipelines orchestrate complete workflows:\n\n| Pipeline | Purpose |\n|----------|---------|\n| `StableDiffusionPipeline` | Text-to-image (SD 1.x\u002F2.x) |\n| `StableDiffusionXLPipeline` | Text-to-image (SDXL) |\n| `StableDiffusion3Pipeline` | Text-to-image (SD 3.0) |\n| `FluxPipeline` | Text-to-image (Flux models) |\n| `StableDiffusionImg2ImgPipeline` | Image-to-image |\n| `StableDiffusionInpaintPipeline` | Inpainting |\n\n### Schedulers\n\nSchedulers control the denoising process:\n\n| Scheduler | Steps | Quality | Use Case |\n|-----------|-------|---------|----------|\n| `EulerDiscreteScheduler` | 20-50 | Good | Default choice |\n| `EulerAncestralDiscreteScheduler` | 20-50 | Good | More variation |\n| `DPMSolverMultistepScheduler` | 15-25 | Excellent | Fast, high quality |\n| `DDIMScheduler` | 50-100 | Good | Deterministic |\n| `LCMScheduler` | 4-8 | Good | Very fast |\n| `UniPCMultistepScheduler` | 15-25 | Excellent | Fast convergence |\n\n### Swapping schedulers\n\n```python\nfrom diffusers import DPMSolverMultistepScheduler\n\n# Swap for faster generation\npipe.scheduler = DPMSolverMultistepScheduler.from_config(\n    pipe.scheduler.config\n)\n\n# Now generate with fewer steps\nimage = pipe(prompt, num_inference_steps=20).images[0]\n```\n\n## Generation parameters\n\n### Key parameters\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `prompt` | Required | Text description of desired image |\n| `negative_prompt` | None | What to avoid in the image |\n| `num_inference_steps` | 50 | Denoising steps (more = better quality) |\n| `guidance_scale` | 7.5 | Prompt adherence (7-12 typical) |\n| `height`, `width` | 512\u002F1024 | Output dimensions (multiples of 8) |\n| `generator` | None | Torch generator for reproducibility |\n| `num_images_per_prompt` | 1 | Batch size |\n\n### Reproducible generation\n\n```python\nimport torch\n\ngenerator = torch.Generator(device=\"cuda\").manual_seed(42)\n\nimage = pipe(\n    prompt=\"A cat wearing a top hat\",\n    generator=generator,\n    num_inference_steps=50\n).images[0]\n```\n\n### Negative prompts\n\n```python\nimage = pipe(\n    prompt=\"Professional photo of a dog in a garden\",\n    negative_prompt=\"blurry, low quality, distorted, ugly, bad anatomy\",\n    guidance_scale=7.5\n).images[0]\n```\n\n## Image-to-image\n\nTransform existing images with text guidance:\n\n```python\nfrom diffusers import AutoPipelineForImage2Image\nfrom PIL import Image\n\npipe = AutoPipelineForImage2Image.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\ninit_image = Image.open(\"input.jpg\").resize((512, 512))\n\nimage = pipe(\n    prompt=\"A watercolor painting of the scene\",\n    image=init_image,\n    strength=0.75,  # How much to transform (0-1)\n    num_inference_steps=50\n).images[0]\n```\n\n## Inpainting\n\nFill masked regions:\n\n```python\nfrom diffusers import AutoPipelineForInpainting\nfrom PIL import Image\n\npipe = AutoPipelineForInpainting.from_pretrained(\n    \"runwayml\u002Fstable-diffusion-inpainting\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\nimage = Image.open(\"photo.jpg\")\nmask = Image.open(\"mask.png\")  # White = inpaint region\n\nresult = pipe(\n    prompt=\"A red car parked on the street\",\n    image=image,\n    mask_image=mask,\n    num_inference_steps=50\n).images[0]\n```\n\n## ControlNet\n\nAdd spatial conditioning for precise control:\n\n```python\nfrom diffusers import StableDiffusionControlNetPipeline, ControlNetModel\nimport torch\n\n# Load ControlNet for edge conditioning\ncontrolnet = ControlNetModel.from_pretrained(\n    \"lllyasviel\u002Fcontrol_v11p_sd15_canny\",\n    torch_dtype=torch.float16\n)\n\npipe = StableDiffusionControlNetPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    controlnet=controlnet,\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Use Canny edge image as control\ncontrol_image = get_canny_image(input_image)\n\nimage = pipe(\n    prompt=\"A beautiful house in the style of Van Gogh\",\n    image=control_image,\n    num_inference_steps=30\n).images[0]\n```\n\n### Available ControlNets\n\n| ControlNet | Input Type | Use Case |\n|------------|------------|----------|\n| `canny` | Edge maps | Preserve structure |\n| `openpose` | Pose skeletons | Human poses |\n| `depth` | Depth maps | 3D-aware generation |\n| `normal` | Normal maps | Surface details |\n| `mlsd` | Line segments | Architectural lines |\n| `scribble` | Rough sketches | Sketch-to-image |\n\n## LoRA adapters\n\nLoad fine-tuned style adapters:\n\n```python\nfrom diffusers import DiffusionPipeline\n\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Load LoRA weights\npipe.load_lora_weights(\"path\u002Fto\u002Flora\", weight_name=\"style.safetensors\")\n\n# Generate with LoRA style\nimage = pipe(\"A portrait in the trained style\").images[0]\n\n# Adjust LoRA strength\npipe.fuse_lora(lora_scale=0.8)\n\n# Unload LoRA\npipe.unload_lora_weights()\n```\n\n### Multiple LoRAs\n\n```python\n# Load multiple LoRAs\npipe.load_lora_weights(\"lora1\", adapter_name=\"style\")\npipe.load_lora_weights(\"lora2\", adapter_name=\"character\")\n\n# Set weights for each\npipe.set_adapters([\"style\", \"character\"], adapter_weights=[0.7, 0.5])\n\nimage = pipe(\"A portrait\").images[0]\n```\n\n## Memory optimization\n\n### Enable CPU offloading\n\n```python\n# Model CPU offload - moves models to CPU when not in use\npipe.enable_model_cpu_offload()\n\n# Sequential CPU offload - more aggressive, slower\npipe.enable_sequential_cpu_offload()\n```\n\n### Attention slicing\n\n```python\n# Reduce memory by computing attention in chunks\npipe.enable_attention_slicing()\n\n# Or specific chunk size\npipe.enable_attention_slicing(\"max\")\n```\n\n### xFormers memory-efficient attention\n\n```python\n# Requires xformers package\npipe.enable_xformers_memory_efficient_attention()\n```\n\n### VAE slicing for large images\n\n```python\n# Decode latents in tiles for large images\npipe.enable_vae_slicing()\npipe.enable_vae_tiling()\n```\n\n## Model variants\n\n### Loading different precisions\n\n```python\n# FP16 (recommended for GPU)\npipe = DiffusionPipeline.from_pretrained(\n    \"model-id\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\n\n# BF16 (better precision, requires Ampere+ GPU)\npipe = DiffusionPipeline.from_pretrained(\n    \"model-id\",\n    torch_dtype=torch.bfloat16\n)\n```\n\n### Loading specific components\n\n```python\nfrom diffusers import UNet2DConditionModel, AutoencoderKL\n\n# Load custom VAE\nvae = AutoencoderKL.from_pretrained(\"stabilityai\u002Fsd-vae-ft-mse\")\n\n# Use with pipeline\npipe = DiffusionPipeline.from_pretrained(\n    \"stable-diffusion-v1-5\u002Fstable-diffusion-v1-5\",\n    vae=vae,\n    torch_dtype=torch.float16\n)\n```\n\n## Batch generation\n\nGenerate multiple images efficiently:\n\n```python\n# Multiple prompts\nprompts = [\n    \"A cat playing piano\",\n    \"A dog reading a book\",\n    \"A bird painting a picture\"\n]\n\nimages = pipe(prompts, num_inference_steps=30).images\n\n# Multiple images per prompt\nimages = pipe(\n    \"A beautiful sunset\",\n    num_images_per_prompt=4,\n    num_inference_steps=30\n).images\n```\n\n## Common workflows\n\n### Workflow 1: High-quality generation\n\n```python\nfrom diffusers import StableDiffusionXLPipeline, DPMSolverMultistepScheduler\nimport torch\n\n# 1. Load SDXL with optimizations\npipe = StableDiffusionXLPipeline.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16,\n    variant=\"fp16\"\n)\npipe.to(\"cuda\")\npipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)\npipe.enable_model_cpu_offload()\n\n# 2. Generate with quality settings\nimage = pipe(\n    prompt=\"A majestic lion in the savanna, golden hour lighting, 8k, detailed fur\",\n    negative_prompt=\"blurry, low quality, cartoon, anime, sketch\",\n    num_inference_steps=30,\n    guidance_scale=7.5,\n    height=1024,\n    width=1024\n).images[0]\n```\n\n### Workflow 2: Fast prototyping\n\n```python\nfrom diffusers import AutoPipelineForText2Image, LCMScheduler\nimport torch\n\n# Use LCM for 4-8 step generation\npipe = AutoPipelineForText2Image.from_pretrained(\n    \"stabilityai\u002Fstable-diffusion-xl-base-1.0\",\n    torch_dtype=torch.float16\n).to(\"cuda\")\n\n# Load LCM LoRA for fast generation\npipe.load_lora_weights(\"latent-consistency\u002Flcm-lora-sdxl\")\npipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)\npipe.fuse_lora()\n\n# Generate in ~1 second\nimage = pipe(\n    \"A beautiful landscape\",\n    num_inference_steps=4,\n    guidance_scale=1.0\n).images[0]\n```\n\n## Common issues\n\n**CUDA out of memory:**\n```python\n# Enable memory optimizations\npipe.enable_model_cpu_offload()\npipe.enable_attention_slicing()\npipe.enable_vae_slicing()\n\n# Or use lower precision\npipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)\n```\n\n**Black\u002Fnoise images:**\n```python\n# Check VAE configuration\n# Use safety checker bypass if needed\npipe.safety_checker = None\n\n# Ensure proper dtype consistency\npipe = pipe.to(dtype=torch.float16)\n```\n\n**Slow generation:**\n```python\n# Use faster scheduler\nfrom diffusers import DPMSolverMultistepScheduler\npipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)\n\n# Reduce steps\nimage = pipe(prompt, num_inference_steps=20).images[0]\n```\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Custom pipelines, fine-tuning, deployment\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fdiffusers\n- **Repository**: https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fdiffusers\n- **Model Hub**: https:\u002F\u002Fhuggingface.co\u002Fmodels?library=diffusers\n- **Discord**: https:\u002F\u002Fdiscord.gg\u002Fdiffusers\n",{"name":5,"tags":24,"author":31,"license":15,"version":32,"description":7,"dependencies":33},[25,26,27,28,29,30],"Image Generation","Stable Diffusion","Diffusers","Text-to-Image","Multimodal","Computer Vision","Orchestra Research","1.0.0",[34,35,36,37],"diffusers>=0.30.0","transformers>=4.41.0","accelerate>=0.31.0","torch>=2.0.0","master","c247302c836de6438d746b45670990eb71691166","top100-runs\u002F084-stable-diffusion",1,"https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool","https:\u002F\u002Fgithub.com\u002Fqhjqhj00","chien.io","2026-05-11 15:07:59",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add qhjqhj00\u002Fresearch-skills-pool --skill stable-diffusion-image-generation",{"tool":52,"command":53},"bunx","bunx skills add qhjqhj00\u002Fresearch-skills-pool --skill stable-diffusion-image-generation",{"tool":55,"command":56},"pnpm","pnpm dlx skills add qhjqhj00\u002Fresearch-skills-pool --skill stable-diffusion-image-generation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002FSKILL.md",true,[60,66,72,76],{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"run.py","SCRIPT","脚本",3425,"https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002Frun.py",{"path":67,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":70,"sourceHtmlUrl":71},"status.json","REFERENCE","参考文档",409,"https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002Fstatus.json",{"path":73,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":74,"sourceHtmlUrl":75},"stderr.txt",675,"https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002Fstderr.txt",{"path":77,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":78,"sourceHtmlUrl":79},"stdout.txt",63,"https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F084-stable-diffusion\u002Fstdout.txt",[81,88,95,102,109,116],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M73JJB6EQPFPPVQK2Y9N67Q0","transformer-lens-interpretability","Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F130-transformer-lens\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":19},"01M73JHTANRRWEE9HP98TPEHAY","wiki-vitepress","Packages generated wiki Markdown into a VitePress static site with dark theme, dark-mode Mermaid diagrams with click-to-zoom, and production build output. Use when the user wants to create a browsable website from generated wiki pages.","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F126-wiki-vitepress\u002FSKILL.md",[],[],{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":19},"01M73JHJBMVFW46F1Y8K3SBF43","phoenix-observability","Open-source AI observability platform for LLM tracing, evaluation, and monitoring. Use when debugging LLM applications with detailed traces, running evaluations on datasets, or monitoring production AI systems with real-time insights.","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F124-phoenix\u002FSKILL.md",[],[],{"skillNo":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":19},"01M73JHFPA3A2806PMP42ZN86Q","frontend-ui-dark-ts","Build dark-themed React applications using Tailwind CSS with custom theming, glassmorphism effects, and Framer Motion animations. Use when creating dashboards, admin panels, or data-rich interfaces with a refined dark aesthetic.","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F123-frontend-ui-dark-ts\u002FSKILL.md",[],[],{"skillNo":110,"name":111,"nameZh":6,"description":112,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":113,"categories":114,"keywords":115,"publishedAt":19},"01M73JFVBVK57NRJS49TT2VDCT","tensorboard","Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F113-tensorboard\u002FSKILL.md",[],[],{"skillNo":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":120,"licenseFileUrl":6,"sourceHtmlUrl":121,"categories":122,"keywords":123,"publishedAt":19},"01M73JFF6PK6ZCN41C89J7HS2B","markdown-mermaid-writing","Comprehensive markdown and Mermaid diagram writing skill. Use when creating any scientific document, report, analysis, or visualization. Establishes text-based diagrams as the default documentation standard with full style guides (markdown + mermaid), 24 diagram type references, and 9 document templates.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fqhjqhj00\u002Fresearch-skills-pool\u002Fblob\u002Fmaster\u002Ftop100-runs\u002F110-markdown-mermaid-writing\u002FSKILL.md",[],[]]