---

title: Stable Diffusion date: 2024-12-01 10:00:00 background: bg-[#A855F7] tags: - AI - Image - Open Source - Generation categories: - AI intro: Stable Diffusion cheat sheet with WebUI usage, prompt techniques, samplers and model configuration plugins: - copyCode

Overview

What is Stable Diffusion

• Open source AI image generation model
• Can run locally
• Supports multiple models and extensions
• Highly customizable

Main UI Tools

Main Models

Prompt Basics

Basic Format

[subject], [description], [environment], [style], [quality]

Positive Prompt Example

1girl, long hair, blue eyes,
white dress, standing in garden,
detailed background, soft lighting,
masterpiece, best quality, 8k

Negative Prompt Example

lowres, bad anatomy, bad hands,
text, error, missing fingers,
extra digit, fewer digits, cropped,
worst quality, low quality,
normal quality, jpeg artifacts,
signature, watermark, username, blurry

Weight Syntax

Increase weight

(beautiful:1.3)
((important))        # ~1.21x
(((very important))) # ~1.33x

Decrease weight

(unwanted:0.7)
[less important]     # ~0.9x

Prompt Blending

AND association

cat AND dog        # Generate both cat and dog

BREAK separation

portrait BREAK landscape

Generation Parameters

Samplers {.row-span-2}

Sampling Steps

CFG Scale

Resolution

SD 1.5

512x512   # Default
512x768   # Portrait
768x512   # Landscape

SDXL

1024x1024 # Default
1024x1536 # Portrait
1536x1024 # Landscape

Seed

-1        # Random seed
Fixed     # Reproducible results

Models & LoRA

Checkpoint Models

Realistic

Realistic Vision
SDXL Realistic
Juggernaut XL

Anime

Anything V5
Counterfeit
Animagine XL

LoRA Usage

Basic format

<lora:lora_name:weight>

Example

1girl, <lora:koreanDollLikeness:0.7>,
detailed face, masterpiece

Multiple LoRAs

portrait, <lora:styleA:0.6>,
<lora:styleB:0.4>, best quality

Embeddings (Textual Inversion)

Negative embeddings

(in negative prompt)
EasyNegative, badhandv4, ng_deepnegative

VAE

Recommended

vae-ft-mse-840000
sdxl_vae

Purpose

Improve colors and details
Reduce grayish appearance

Common Features

Hires. Fix

Settings

Upscaler: R-ESRGAN 4x+
Hires steps: 10-20
Denoising strength: 0.3-0.5
Upscale by: 1.5-2

Inpainting

Steps

1. Upload image to img2img
2. Select Inpaint
3. Draw mask area
4. Set Denoising 0.4-0.7
5. Generate

ControlNet

Common Preprocessors

Canny         # Edge detection
OpenPose      # Pose control
Depth         # Depth map
Lineart       # Line art
Tile          # Detail enhancement
IP-Adapter    # Image reference

img2img

Parameters

Denoising strength:
0.1-0.3  # Minor changes
0.4-0.6  # Medium changes
0.7-0.9  # Major changes

Quality Prompts

Quality Positive

masterpiece, best quality,
ultra detailed, 8k uhd,
high resolution, intricate details,
sharp focus, photorealistic

Quality Negative

worst quality, low quality,
normal quality, lowres,
bad anatomy, bad hands,
text, error, cropped,
jpeg artifacts, signature,
watermark, username

SDXL Recommended

Positive

score_9, score_8_up, score_7_up
detailed, masterpiece

Negative

score_6, score_5, score_4,
low quality, blurry

Prompt Templates {.cols-2}

Portrait

Template

Positive:
1girl, solo, portrait, looking at viewer,
[hair color] hair, [hairstyle], [eye color] eyes,
[clothing], detailed face, beautiful eyes,
natural lighting, soft shadows,
masterpiece, best quality, 8k

Negative:
lowres, bad anatomy, bad hands, text,
error, missing fingers, extra digit,
fewer digits, cropped, worst quality,
low quality, normal quality

Example

1girl, solo, portrait, looking at viewer,
blonde hair, long wavy hair, green eyes,
white blouse, detailed face, beautiful eyes,
natural lighting, soft shadows,
masterpiece, best quality, 8k

Landscape

Template

Positive:
landscape, [scene type],
[time] sky, [weather],
volumetric lighting, atmospheric,
detailed environment, high detail,
masterpiece, best quality, 8k

Negative:
blurry, low quality, text, watermark,
ugly, deformed, oversaturated

Example

landscape, mountain lake at sunset,
orange sky, scattered clouds,
volumetric lighting, atmospheric,
detailed environment, high detail,
masterpiece, best quality, 8k

Anime Character

Template

Positive:
1girl, anime style, [character description],
[clothing], [pose/action],
detailed, vibrant colors,
masterpiece, best quality

Negative:
lowres, bad anatomy, bad hands,
worst quality, low quality,
extra limbs, missing limbs

Example

1girl, anime style, silver hair, red eyes,
black gothic dress, holding umbrella,
standing in rain, detailed, vibrant colors,
masterpiece, best quality

Product Photo

Template

Positive:
product photography, [product],
studio lighting, white background,
professional, commercial,
high detail, 8k uhd

Negative:
blurry, low quality, shadows,
text, watermark, cluttered

More Prompt Examples

Fantasy Characters

Wizard

old wizard with long white beard,
wearing purple robes with golden stars,
holding glowing staff, magical aura,
fantasy art style, detailed,
dramatic lighting, masterpiece

Warrior

female warrior in ornate armor,
battle stance, sword drawn,
wind blowing through hair,
epic fantasy, cinematic lighting,
detailed armor, masterpiece, 8k

Sci-Fi Scenes

Space Station

interior of futuristic space station,
large observation window showing Earth,
clean white surfaces, holographic displays,
sci-fi movie style, cinematic,
volumetric lighting, 8k

Robot

advanced humanoid robot,
sleek chrome design, glowing blue eyes,
standing in laboratory,
sci-fi concept art, highly detailed,
studio lighting, 8k

Nature & Animals

Wildlife

majestic lion in savanna,
golden hour lighting, dramatic sky,
wildlife photography style,
sharp focus, detailed fur,
National Geographic quality

Underwater

coral reef teeming with tropical fish,
sunlight rays through water,
underwater photography, vivid colors,
sharp focus, detailed, 8k

Food Photography

Dessert

chocolate lava cake with molten center,
fresh raspberries, mint leaf garnish,
dark slate plate, food photography,
shallow depth of field, studio lighting,
professional, appetizing

Asian Cuisine

bowl of ramen with soft-boiled egg,
steam rising, chopsticks, detailed noodles,
overhead shot, food photography style,
warm lighting, professional, 8k

Architecture

Modern Interior

minimalist living room interior,
floor to ceiling windows, city view,
neutral colors, designer furniture,
architectural photography, natural light,
clean lines, interior design magazine

Historical

gothic cathedral interior,
sunlight streaming through stained glass,
stone pillars, vaulted ceiling,
architectural photography, dramatic lighting,
detailed stonework, 8k

WebUI Extensions

Recommended Extensions

Image Enhancement

Ultimate SD Upscale
Tiled Diffusion
ADetailer (face fix)

Prompt Tools

Dynamic Prompts
Prompt All-in-one

ControlNet

sd-webui-controlnet

ADetailer Settings

Model: face_yolov8n.pt
Confidence: 0.3
Denoising: 0.4

Performance Optimization

VRAM Optimization

Command line args

--medvram           # Medium VRAM
--lowvram           # Low VRAM
--xformers          # VRAM optimization
--opt-sdp-attention # PyTorch 2.0

Speed Settings

--api               # Enable API
--no-half-vae       # Avoid black images
--precision full    # Precision issues

Recommended Config

Troubleshooting

Gray/Dark Images

Solutions

1. Change VAE
2. Use --no-half-vae
3. Adjust CFG to 7-8

Hand Issues

Solutions

Negative: bad hands, extra fingers
Use ADetailer
Use ControlNet OpenPose

Face Issues

Solutions

Use ADetailer extension
Use face_yolov8n model
Add "detailed face" to prompt

Slow Generation

Solutions

Lower resolution
Reduce steps to 20-25
Use UniPC sampler
Enable xformers

ComfyUI Basics

Basic Nodes

Load Checkpoint  # Load model
CLIP Text Encode # Text encoding
KSampler         # Sampler
VAE Decode       # VAE decode
Save Image       # Save image

Basic Workflow

Checkpoint → CLIP → KSampler → VAE Decode → Save
                ↑
            Empty Latent

Advantages

• Node-based workflow
• Highly customizable
• Better memory management
• Complex workflow support