title: “Stable Diffusion — Text-to-image generation, inpainting, and img2img” sidebar_label: “Stable Diffusion” description: “Text-to-image generation, inpainting, and img2img”

{/* 本页面由 website/scripts/generate-skill-docs.py 根据技能对应的 SKILL.md 文件自动生成。请直接编辑源文件 SKILL.md，而非此页面。 */}

Stable Diffusion

文本转图像生成、图像修复以及图像间转换功能。

技能元数据

参考：完整 SKILL.md 文件

:::info 以下是当触发该技能时 Hermes 所加载的完整技能定义。技能激活后，智能体将看到这些内容作为操作指令。

::

使用 Stable Diffusion 进行图像生成

介绍如何利用 HuggingFace Diffusers 库通过 Stable Diffusion 生成图像。

何时使用 Stable Diffusion

以下情况适合使用 Stable Diffusion：

• 根据文本描述生成图像
• 执行图像间转换（风格迁移、图像增强）
• 图像修复（填充被遮盖的区域）
• 图像扩展（将图像内容延伸至边界之外）
• 创建现有图像的变体
• 构建自定义的图像生成工作流

主要功能：

• 文本生成图像：根据自然语言描述生成图片
• 图像转图像：在文字引导下对现有图像进行改造
• 图像修复：用符合上下文的内容填充被遮盖的区域
• ControlNet：添加空间控制功能（边缘、姿态、深度）
• LoRA支持：实现高效微调与风格适配
• 多种模型支持：兼容SD 1.5、SDXL、SD 3.0及Flux模型

可选替代方案：

• DALL-E 3：适用于无需GPU的基于API的图像生成
• Midjourney：适合生成具有艺术风格的作品
• Imagen：适用于与Google Cloud集成
• Leonardo.ai：适合基于网页的创意工作流程

快速入门

安装

pip install diffusers transformers accelerate torch
pip install xformers  # Optional: memory-efficient attention

基础文本转图像功能

from diffusers import DiffusionPipeline
import torch

# Load pipeline (auto-detects model type)
pipe = DiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16
)
pipe.to("cuda")

# Generate image
image = pipe(
    "A serene mountain landscape at sunset, highly detailed",
    num_inference_steps=50,
    guidance_scale=7.5
).images[0]

image.save("output.png")

使用 SDXL（更高画质）

from diffusers import AutoPipelineForText2Image
import torch

pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16"
)
pipe.to("cuda")

# Enable memory optimization
pipe.enable_model_cpu_offload()

image = pipe(
    prompt="A futuristic city with flying cars, cinematic lighting",
    height=1024,
    width=1024,
    num_inference_steps=30
).images[0]

架构概览

三大核心组件设计

Diffusers 由三个核心组件构成：

Pipeline (orchestration)
├── Model (neural networks)
│   ├── UNet / Transformer (noise prediction)
│   ├── VAE (latent encoding/decoding)
│   └── Text Encoder (CLIP/T5)
└── Scheduler (denoising algorithm)

流水线推理流程

需对完整的输入内容进行翻译，不得提前终止处理。

Text Prompt → Text Encoder → Text Embeddings
                                    ↓
Random Noise → [Denoising Loop] ← Scheduler
                      ↓
               Predicted Noise
                      ↓
              VAE Decoder → Final Image

核心概念

流水线

流水线用于统筹整个工作流程：

调度器

调度器用于控制去噪过程：

更换调度器

from diffusers import DPMSolverMultistepScheduler

# Swap for faster generation
pipe.scheduler = DPMSolverMultistepScheduler.from_config(
    pipe.scheduler.config
)

# Now generate with fewer steps
image = pipe(prompt, num_inference_steps=20).images[0]

生成参数

核心参数

可复现的图像生成

import torch

generator = torch.Generator(device="cuda").manual_seed(42)

image = pipe(
    prompt="A cat wearing a top hat",
    generator=generator,
    num_inference_steps=50
).images[0]

负面提示词

image = pipe(
    prompt="Professional photo of a dog in a garden",
    negative_prompt="blurry, low quality, distorted, ugly, bad anatomy",
    guidance_scale=7.5
).images[0]

图像到图像转换

在文本引导下对现有图像进行变换：

from diffusers import AutoPipelineForImage2Image
from PIL import Image

pipe = AutoPipelineForImage2Image.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16
).to("cuda")

init_image = Image.open("input.jpg").resize((512, 512))

image = pipe(
    prompt="A watercolor painting of the scene",
    image=init_image,
    strength=0.75,  # How much to transform (0-1)
    num_inference_steps=50
).images[0]

修复填充功能

填充被遮盖的区域：

from diffusers import AutoPipelineForInpainting
from PIL import Image

pipe = AutoPipelineForInpainting.from_pretrained(
    "runwayml/stable-diffusion-inpainting",
    torch_dtype=torch.float16
).to("cuda")

image = Image.open("photo.jpg")
mask = Image.open("mask.png")  # White = inpaint region

result = pipe(
    prompt="A red car parked on the street",
    image=image,
    mask_image=mask,
    num_inference_steps=50
).images[0]

ControlNet

通过空间约束实现精准控制：

from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
import torch

# Load ControlNet for edge conditioning
controlnet = ControlNetModel.from_pretrained(
    "lllyasviel/control_v11p_sd15_canny",
    torch_dtype=torch.float16
)

pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    controlnet=controlnet,
    torch_dtype=torch.float16
).to("cuda")

# Use Canny edge image as control
control_image = get_canny_image(input_image)

image = pipe(
    prompt="A beautiful house in the style of Van Gogh",
    image=control_image,
    num_inference_steps=30
).images[0]

可用的 ControlNets

LoRA 适配器

加载经过微调的风格适配器：

from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16
).to("cuda")

# Load LoRA weights
pipe.load_lora_weights("path/to/lora", weight_name="style.safetensors")

# Generate with LoRA style
image = pipe("A portrait in the trained style").images[0]

# Adjust LoRA strength
pipe.fuse_lora(lora_scale=0.8)

# Unload LoRA
pipe.unload_lora_weights()

多个LoRA模型

# Load multiple LoRAs
pipe.load_lora_weights("lora1", adapter_name="style")
pipe.load_lora_weights("lora2", adapter_name="character")

# Set weights for each
pipe.set_adapters(["style", "character"], adapter_weights=[0.7, 0.5])

image = pipe("A portrait").images[0]

内存优化

启用 CPU 卸载功能

# Model CPU offload - moves models to CPU when not in use
pipe.enable_model_cpu_offload()

# Sequential CPU offload - more aggressive, slower
pipe.enable_sequential_cpu_offload()

注意力切片

# Reduce memory by computing attention in chunks
pipe.enable_attention_slicing()

# Or specific chunk size
pipe.enable_attention_slicing("max")

xFormers高效内存注意力机制

# Requires xformers package
pipe.enable_xformers_memory_efficient_attention()

大尺寸图像的VAE切片功能

# Decode latents in tiles for large images
pipe.enable_vae_slicing()
pipe.enable_vae_tiling()

模型版本

加载不同精度版本

# FP16 (recommended for GPU)
pipe = DiffusionPipeline.from_pretrained(
    "model-id",
    torch_dtype=torch.float16,
    variant="fp16"
)

# BF16 (better precision, requires Ampere+ GPU)
pipe = DiffusionPipeline.from_pretrained(
    "model-id",
    torch_dtype=torch.bfloat16
)

加载特定组件

from diffusers import UNet2DConditionModel, AutoencoderKL

# Load custom VAE
vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse")

# Use with pipeline
pipe = DiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    vae=vae,
    torch_dtype=torch.float16
)

批量生成

高效地创建多张图像：

# Multiple prompts
prompts = [
    "A cat playing piano",
    "A dog reading a book",
    "A bird painting a picture"
]

images = pipe(prompts, num_inference_steps=30).images

# Multiple images per prompt
images = pipe(
    "A beautiful sunset",
    num_images_per_prompt=4,
    num_inference_steps=30
).images

常见工作流程

工作流程 1：高质量生成

from diffusers import StableDiffusionXLPipeline, DPMSolverMultistepScheduler
import torch

# 1. Load SDXL with optimizations
pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16"
)
pipe.to("cuda")
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_model_cpu_offload()

# 2. Generate with quality settings
image = pipe(
    prompt="A majestic lion in the savanna, golden hour lighting, 8k, detailed fur",
    negative_prompt="blurry, low quality, cartoon, anime, sketch",
    num_inference_steps=30,
    guidance_scale=7.5,
    height=1024,
    width=1024
).images[0]

工作流 2：快速原型设计

from diffusers import AutoPipelineForText2Image, LCMScheduler
import torch

# Use LCM for 4-8 step generation
pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16
).to("cuda")

# Load LCM LoRA for fast generation
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
pipe.fuse_lora()

# Generate in ~1 second
image = pipe(
    "A beautiful landscape",
    num_inference_steps=4,
    guidance_scale=1.0
).images[0]

常见问题

CUDA 内存不足：

# Enable memory optimizations
pipe.enable_model_cpu_offload()
pipe.enable_attention_slicing()
pipe.enable_vae_slicing()

# Or use lower precision
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)

黑色/噪点图像：

# Check VAE configuration
# Use safety checker bypass if needed
pipe.safety_checker = None

# Ensure proper dtype consistency
pipe = pipe.to(dtype=torch.float16)

生成速度过慢：

# Use faster scheduler
from diffusers import DPMSolverMultistepScheduler
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

# Reduce steps
image = pipe(prompt, num_inference_steps=20).images[0]

参考资料

• 高级用法 - 自定义流程、微调及部署方法
• 故障排查 - 常见问题与解决方案

资源链接

• 文档：https://huggingface.co/docs/diffusers
• 代码仓库：https://github.com/huggingface/diffusers
• 模型中心：https://huggingface.co/models?library=diffusers
• Discord社区：https://discord.gg/diffusers