AI绘图工具指南 - Stable Diffusion与Midjourney
全面掌握主流AI绘图工具，创作高质量AI艺术作品
前置知识：了解基本的AI概念
本文重点：主流AI绘图工具使用与提示词技巧
一、AI绘图概述
1.1 技术原理
AI绘图核心技术：扩散模型 (Diffusion Model)
工作原理：
1. 前向扩散：逐步向图像添加噪声
2. 反向去噪：学习从噪声恢复图像
3. 条件引导：通过文本提示控制生成
主流模型：
├── Stable Diffusion: 开源、可本地部署
├── Midjourney: 托管服务、艺术效果强
├── DALL-E 3: OpenAI、理解能力强
└── Adobe Firefly: 商业安全、版权友好1.2 工具对比
AI绘图工具对比：
┌─────────────┬──────────┬──────────┬──────────┬──────────┐
│ 特性 │ SD │ MJ │ DALL-E │ Firefly │
├─────────────┼──────────┼──────────┼──────────┼──────────┤
│ 开源免费 │ ✓ │ ✗ │ ✗ │ ✗ │
│ 本地部署 │ ✓ │ ✗ │ ✗ │ ✗ │
│ 商业使用 │ ✓ │ 付费版 │ 付费 │ ✓ │
│ 学习曲线 │ 陡峭 │ 平缓 │ 平缓 │ 平缓 │
│ 控制精细度 │ 高 │ 中 │ 中 │ 中 │
│ 艺术风格 │ 多样 │ 强 │ 中 │ 中 │
│ 中文支持 │ 一般 │ 弱 │ 强 │ 强 │
└─────────────┴──────────┴──────────┴──────────┴──────────┘二、Stable Diffusion
2.1 本地部署
# 方案一：使用Automatic1111 WebUI（推荐）
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui
./webui.sh
# 方案二：使用ComfyUI（节点式工作流）
git clone https://github.com/comfyanonymous/ComfyUI.git
# 按照README安装依赖
# 方案三：使用Diffusers库（Python）
pip install diffusers transformers accelerate2.2 Python API使用
from diffusers import StableDiffusionPipeline
import torch
# 加载模型
model_id = "runwayml/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(
model_id,
torch_dtype=torch.float16,
safety_checker=None
)
pipe = pipe.to("cuda")
# 生成图像
prompt = "a beautiful sunset over the ocean, highly detailed, 4k"
negative_prompt = "blurry, low quality, distorted"
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
num_inference_steps=30,
guidance_scale=7.5,
width=512,
height=512
).images[0]
image.save("output.png")2.3 提示词技巧
"""
Stable Diffusion提示词结构：
[主体描述] + [风格] + [质量词] + [技术参数]
示例：
- 主体：a young woman with long hair
- 风格：in the style of anime, studio ghibli
- 质量：masterpiece, best quality, highly detailed
- 技术：8k uhd, sharp focus, cinematic lighting
负向提示词（避免出现的内容）：
- blurry, low quality, distorted
- bad anatomy, extra fingers, missing limbs
- watermark, text, signature
"""
# 高质量提示词模板
quality_prompts = {
"realistic": """
masterpiece, best quality, ultra realistic,
8k uhd, photorealistic, sharp focus,
cinematic lighting, highly detailed,
professional photography
""",
"anime": """
masterpiece, best quality, anime style,
highly detailed, vibrant colors,
studio ghibli style, makoto shinkai
""",
"fantasy": """
masterpiece, best quality, fantasy art,
digital painting, concept art,
trending on artstation, detailed
"""
}
negative_prompts = """
low quality, bad quality, sketches,
bad anatomy, deformed, disfigured,
extra limbs, missing limbs,
floating limbs, disconnected limbs,
mutation, mutated, ugly, disgusting,
blurry, amputation, watermark, text
"""
def generate_with_style(prompt, style="realistic"):
"""带风格模板的生成"""
full_prompt = f"{prompt}, {quality_prompts[style]}"
image = pipe(
prompt=full_prompt,
negative_prompt=negative_prompts,
num_inference_steps=30,
guidance_scale=7.5
).images[0]
return image2.4 高级功能
# 1. 图生图 (Image-to-Image)
from diffusers import StableDiffusionImg2ImgPipeline
from PIL import Image
img2img = StableDiffusionImg2ImgPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
torch_dtype=torch.float16
).to("cuda")
init_image = Image.open("input.png").convert("RGB")
init_image = init_image.resize((512, 512))
image = img2img(
prompt="a oil painting of the image",
image=init_image,
strength=0.75, # 变换强度 0-1
guidance_scale=7.5
).images[0]
# 2. 局部重绘 (Inpainting)
from diffusers import StableDiffusionInpaintPipeline
inpaint = StableDiffusionInpaintPipeline.from_pretrained(
"runwayml/stable-diffusion-inpainting",
torch_dtype=torch.float16
).to("cuda")
# 原图和掩码图
original = Image.open("original.png")
mask = Image.open("mask.png") # 白色区域将被重绘
result = inpaint(
prompt="a cat sitting on the chair",
image=original,
mask_image=mask,
guidance_scale=7.5
).images[0]
# 3. ControlNet（精确控制）
# pip install controlnet-aux
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
from diffusers.utils import load_image
# 加载ControlNet模型
controlnet = ControlNetModel.from_pretrained(
"lllyasviel/sd-controlnet-canny",
torch_dtype=torch.float16
)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
controlnet=controlnet,
torch_dtype=torch.float16
).to("cuda")
# 准备控制图像（如边缘检测图）
control_image = load_image("control.png")
image = pipe(
prompt="a modern building",
image=control_image,
num_inference_steps=20
).images[0]三、Midjourney
3.1 基本使用
Midjourney通过Discord使用：
基本命令：
/imagine prompt: [描述]
/blend: 混合多张图片
/describe: 根据图片生成描述
/shorten: 优化提示词
参数：
--ar 16:9 设置宽高比
--v 6 使用版本6
--q 2 质量参数(1-2)
--s 750 风格化程度(0-1000)
--c 50 混沌度(0-100)
--seed 12345 随机种子
--no xxx 负向提示词3.2 提示词技巧
Midjourney提示词结构：
[主体] + [环境] + [风格] + [参数]
示例提示词：
1. 人像：
A portrait of a young woman,
soft natural lighting,
bokeh background,
shot on 85mm lens,
cinematic look --ar 2:3 --v 6
2. 风景：
Majestic mountain landscape at sunset,
dramatic clouds,
golden hour lighting,
epic scale,
photorealistic --ar 16:9 --v 6
3. 概念艺术：
A futuristic city with flying cars,
neon lights reflecting on wet streets,
cyberpunk style,
highly detailed --ar 21:9 --v 6
4. 产品渲染：
Smart watch product photography,
minimalist white background,
studio lighting,
professional product shot --ar 1:1 --v 6
风格关键词：
- photorealistic: 照片级真实
- anime: 动漫风格
- oil painting: 油画风格
- watercolor: 水彩画
- 3d render: 3D渲染
- pixel art: 像素艺术
- minimalist: 极简主义3.3 高级技巧
1. 多提示词（权重控制）
cat::2, dog::1 # cat权重是dog的两倍
2. 图像提示
/imagine prompt: https://url1.png https://url2.png
a blend of these images --iw 0.5
3. 平铺图案
/seamless pattern of flowers,
repeating design --tile
4. 角色一致性（使用角色参考）
--cref https://url.png
5. 风格参考
--sref https://style-image.png
6. 负向提示词
beautiful landscape --no people, buildings
7. Niji模式（动漫专用）
/niji prompt: anime girl, detailed四、DALL-E 3
4.1 API使用
from openai import OpenAI
client = OpenAI()
# 生成图像
response = client.images.generate(
model="dall-e-3",
prompt="a futuristic city at night with neon lights",
size="1024x1024",
quality="hd", # standard or hd
n=1,
style="vivid" # vivid or natural
)
image_url = response.data[0].url
print(f"图像URL: {image_url}")
# 图像编辑
response = client.images.edit(
model="dall-e-2",
image=open("original.png", "rb"),
mask=open("mask.png", "rb"),
prompt="add a cat to the image",
n=1,
size="1024x1024"
)
# 变体生成
response = client.images.create_variation(
image=open("original.png", "rb"),
n=3,
size="1024x1024"
)4.2 提示词优化
"""
DALL-E 3提示词特点：
- 自然语言理解强
- 支持复杂描述
- 会自动优化提示词
提示词建议：
1. 描述性强：详细描述场景
2. 指定风格：明确艺术风格
3. 说明用途：如"海报设计"
"""
# 提示词示例
prompts = [
# 插画
"""A whimsical children's book illustration showing
a small mouse having a tea party with a butterfly
in a cozy mushroom house, warm colors, gentle lighting""",
# 产品
"""Professional product photography of a luxury
watch on a dark marble surface, dramatic lighting,
reflecting the watch's gold details, high-end
advertisement style""",
# 概念艺术
"""A digital concept art of a floating island city
with waterfalls cascading over the edges,
bioluminescent plants, fantasy architecture,
epic scale, dramatic sky"""
]五、提示词工程
5.1 通用提示词模板
class ImagePromptBuilder:
"""AI绘图提示词构建器"""
def __init__(self):
self.subject = ""
self.style = ""
self.quality = ""
self.lighting = ""
self.composition = ""
self.negative = ""
def set_subject(self, subject):
self.subject = subject
return self
def set_style(self, style):
style_map = {
"realistic": "photorealistic, ultra realistic, real life",
"anime": "anime style, manga style, cel shading",
"oil_painting": "oil painting, classical art, masterpiece",
"digital_art": "digital art, concept art, digital painting",
"3d": "3d render, octane render, unreal engine",
}
self.style = style_map.get(style, style)
return self
def set_quality(self, level="high"):
quality_map = {
"high": "masterpiece, best quality, highly detailed, 8k",
"medium": "good quality, detailed",
"low": "normal quality"
}
self.quality = quality_map.get(level, level)
return self
def set_lighting(self, lighting):
lighting_map = {
"golden_hour": "golden hour lighting, warm tones",
"blue_hour": "blue hour, cool tones, twilight",
"studio": "studio lighting, soft light, professional",
"dramatic": "dramatic lighting, high contrast, chiaroscuro",
}
self.lighting = lighting_map.get(lighting, lighting)
return self
def set_composition(self, composition):
composition_map = {
"portrait": "portrait shot, bokeh background, 85mm",
"landscape": "landscape shot, wide angle, panoramic",
"closeup": "close-up shot, macro, detailed",
"full_body": "full body shot, fashion photography",
}
self.composition = composition_map.get(composition, composition)
return self
def set_negative(self, negative):
self.negative = negative
return self
def build(self):
prompt_parts = [
self.subject,
self.style,
self.quality,
self.lighting,
self.composition
]
prompt = ", ".join([p for p in prompt_parts if p])
return {
"prompt": prompt,
"negative_prompt": self.negative
}
# 使用示例
builder = ImagePromptBuilder()
result = builder.set_subject("a beautiful woman in a garden") \
.set_style("realistic") \
.set_quality("high") \
.set_lighting("golden_hour") \
.set_composition("portrait") \
.set_negative("blurry, low quality, distorted") \
.build()
print(result["prompt"])5.2 风格参考库
style_library = {
"photography": {
"portrait": "professional portrait photography, bokeh, 85mm lens, studio lighting",
"landscape": "epic landscape photography, wide angle, natural lighting, 16k",
"macro": "macro photography, extreme detail, shallow depth of field",
"street": "street photography, candid, urban, documentary style"
},
"art": {
"oil_painting": "oil painting, classical art, impasto, rich colors",
"watercolor": "watercolor painting, soft edges, flowing colors",
"impressionism": "impressionism style, Claude Monet, visible brushstrokes",
"surrealism": "surrealism, Salvador Dali style, dreamlike, impossible scenes"
},
"digital": {
"concept_art": "concept art, digital painting, artstation trending",
"fantasy": "fantasy art, magical, epic, detailed digital painting",
"scifi": "science fiction art, futuristic, cyberpunk, neon lights",
"anime": "anime style, manga, Japanese animation, vibrant colors"
},
"3d": {
"realistic_3d": "3d render, octane render, unreal engine 5, ray tracing",
"stylized_3d": "stylized 3d, Pixar style, cartoon render",
"architectural": "architectural visualization, interior design, clean lines"
}
}参考资源
- Stable Diffusion WebUI - 最流行的界面
- ComfyUI - 节点式工作流
- Midjourney文档 - 官方指南
- DALL-E API - OpenAI图像API
- Civitai - 模型与作品分享平台
- PromptHero - 提示词搜索库
- ControlNet - 精确控制生成
上一篇：MCP协议详解 返回：AI工具链 最后更新: 2026年4月18日
讨论与反馈