Stable Diffusion Prompting Techniques: Advanced Guide

Stable Diffusion offers unmatched customization for AI image generation. This advanced guide covers professional prompting techniques.

Understanding Stable Diffusion

What Sets It Apart

• Open source: Free to use and modify
• Local deployment: Run on your own hardware
• Extensive ecosystem: Thousands of custom models
• Full control: Every parameter adjustable

Key Concepts

• Checkpoints: Full model files (different styles/capabilities)
• LoRAs: Lightweight model modifications
• Embeddings: Trained concepts for specific elements
• ControlNet: Structural guidance for generations
• VAE: Visual quality encoders

Basic Prompt Structure

Positive Prompt Components

1. Subject description
2. Style and medium
3. Quality tags
4. Technical specifications

Negative Prompt (Equally Important!)

What to exclude from generation.

Example Structure:

Positive: "a beautiful woman, detailed face, professional photography, studio lighting, sharp focus, 8k, masterpiece"

Negative: "ugly, deformed, blurry, low quality, text, watermark, extra limbs, bad anatomy"

Prompt Weighting

Basic Weighting

Use parentheses to increase emphasis:

• (keyword) = 1.1x weight
• ((keyword)) = 1.21x weight
• (((keyword))) = 1.33x weight

Or specify exact weight:

• (keyword:1.5) = 1.5x weight
• (keyword:0.5) = 0.5x weight

Examples:

a woman with (red hair:1.3), wearing a ((blue dress)), in a garden

Negative Weighting

bad anatomy, (deformed hands:1.4), (extra fingers:1.5)

Quality Tags That Work

Universal Quality Boosters

• masterpiece
• best quality
• highly detailed
• sharp focus
• professional
• 8k uhd
• high resolution

Photography Quality

• RAW photo
• photorealistic
• hyperrealistic
• shot on Canon EOS R5
• 85mm lens
• f/1.4 aperture
• studio lighting

Art Quality

• trending on artstation
• award winning
• concept art
• digital painting
• illustration

Negative Prompt Essentials

Standard Negative Set

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

For Portraits

deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers, deformed, distorted, disfigured, poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation

For Landscapes

text, watermark, signature, logo, people, person, humans, buildings (if unwanted), modern elements (for nature)

Working with Models

Checkpoint Selection

Different checkpoints excel at different tasks:

• Realistic Vision: Photorealistic humans
• DreamShaper: Versatile artistic
• Deliberate: Balanced realism/art
• SDXL: Latest base model
• Pony Diffusion: Stylized/anime

LoRA Usage

Add specific styles or characters:

<lora:add_detail:0.8> detailed photograph of a castle

Popular LoRAs:

• add_detail: Enhances detail
• epi_noiseoffset: Better contrast
• GoodHands: Fixes hand issues
• Style-specific LoRAs

Embedding Usage

Trigger words for trained concepts:

EasyNegative, photograph of a woman, detailed face

ControlNet Techniques

What is ControlNet?

ControlNet provides structural guidance:

• Canny edge detection
• Pose detection (OpenPose)
• Depth maps
• Segmentation maps

Use Cases

1. Pose matching: Copy poses from reference images
2. Architectural control: Maintain building structure
3. Face consistency: Keep facial features
4. Composition control: Exact layout matching

Example Workflow

1. Upload reference image
2. Select ControlNet type (e.g., OpenPose)
3. Adjust influence strength (0.5-1.0)
4. Write prompt for style/details
5. Generate with structural guidance

Advanced Techniques

Prompt Blending

Combine concepts with AND:

a forest path AND autumn colors AND mystical atmosphere

Prompt Scheduling

Change prompts during generation:

[winter:spring:0.5] landscape

First 50% uses "winter", then switches to "spring"

Regional Prompting

Apply different prompts to different areas:

• Use Latent Couple extension
• Define regions with masks
• Assign prompts per region

Wildcards

Dynamic prompt variation:

a __color__ __animal__ in a __location__

With files defining options for each wildcard.

Sampler and Settings

Sampler Selection

• DPM++ 2M Karras: Great balance, recommended
• Euler a: Fast, creative
• DPM++ SDE Karras: High quality, slower
• DDIM: Consistent, good for animation

CFG Scale

• 1-5: More creative, less prompt-following
• 7-9: Balanced (recommended)
• 10-15: Strict prompt following
• 15+: Often oversaturated

Steps

• 20-30: Good for most cases
• 40-50: Higher quality, diminishing returns
• Less than 20: Draft quality

Example Prompts

Photorealistic Portrait

Positive:

RAW photo, portrait of a 25 year old woman, natural skin texture, (realistic skin:1.2), detailed eyes, professional photography, studio lighting, softbox, 85mm lens, f/1.8, shallow depth of field, masterpiece, best quality

Negative:

cartoon, painting, illustration, (worst quality, low quality:1.4), (deformed iris, deformed pupils:1.2), bad anatomy, bad hands, bad proportions, ugly, blurry

Fantasy Landscape

Positive:

epic fantasy landscape, floating islands in the sky, waterfalls cascading into clouds, ancient ruins, magical atmosphere, volumetric lighting, god rays, matte painting, concept art, highly detailed, 8k, trending on artstation

Negative:

modern buildings, people, text, watermark, signature, blurry, low quality

Anime Character

Positive:

1girl, solo, long silver hair, blue eyes, detailed face, wearing school uniform, cherry blossom background, soft lighting, anime style, high quality, detailed

Negative:

lowres, bad anatomy, bad hands, error, missing fingers, extra fingers, cropped, worst quality, low quality, jpeg artifacts

Optimization Tips

1. Start with proven prompts: Modify working examples
2. Test one variable at a time: Isolate what works
3. Use X/Y/Z plot: Compare settings systematically
4. Save successful seeds: Reproduce good results
5. Build prompt templates: Standardize your workflow

Troubleshooting Common Issues

Bad Hands

• Add "bad hands" to negative
• Use LoRAs like GoodHands
• Try different samplers
• Use inpainting to fix

Blurry Results

• Increase steps
• Check VAE settings
• Add "sharp focus" to prompt
• Ensure resolution isn't too high for your VRAM

Wrong Style

• Check checkpoint choice
• Adjust LoRA weights
• Review prompt word order
• Use negative prompts for unwanted styles

Conclusion

Stable Diffusion's power lies in customization. Master prompting, explore the ecosystem, and develop your personal workflow.

Pro tip: Use Image to Prompt to analyze Stable Diffusion outputs and reverse-engineer effective prompts!