Stable Diffusion images contain extensive metadata that can reveal your prompts, model settings, and generation parameters. This guide covers the specific metadata embedded by different Stable Diffusion interfaces and how to remove it completely.
What Metadata Does Stable Diffusion Embed?
Unlike commercial AI generators like DALL-E or MidJourney, Stable Diffusion is open-source, meaning the metadata varies significantly based on which interface you use. However, all versions embed data that can identify your images as AI-generated.
Common Stable Diffusion Metadata
Most SD interfaces embed some combination of:
- Prompt: Your positive and negative prompts
- Model checkpoint: Which model file generated the image
- Sampling method: Euler, DPM++, DDIM, etc.
- Steps and CFG scale: Generation parameters
- Seed: The random seed used
- Size: Original generation dimensions
- LoRA/embeddings: Any additional models used
Metadata by Interface
Automatic1111 (AUTOMATIC1111/stable-diffusion-webui)
A1111 is the most popular Stable Diffusion interface and embeds comprehensive metadata in PNG files:
PNG Info Chunk:
parameters: masterpiece, best quality, 1girl...
Negative prompt: bad anatomy, bad hands...
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7
Seed: 1234567890, Size: 512x768
Model hash: abc123def, Model: dreamshaper_8
This data is stored in the PNG's tEXt chunks and can be read by anyone who downloads your image.
ComfyUI
ComfyUI stores its workflow data differently:
- Full workflow JSON: The entire node graph is embedded
- API format: Simplified generation data
- Custom metadata: Node-specific settings
ComfyUI's metadata is more complex because it preserves the entire workflow, including:
- Node connections
- All node parameters
- Model paths (which may reveal your system setup)
- Custom node configurations
Forge/ReForge
These A1111 forks maintain similar metadata formats but may add:
- Forge-specific optimizations used
- Memory management settings
- Custom extension data
InvokeAI
InvokeAI embeds metadata in a structured format:
- Generation parameters in XMP
- Model information
- Canvas/composition data if using the unified canvas
Fooocus
Fooocus typically embeds:
- Simplified prompt information
- Style presets used
- Quality settings
How to Remove Stable Diffusion Metadata
Method 1: AI Metadata Cleaner (Recommended)
Our AI Metadata Cleaner handles all Stable Diffusion interfaces:
- Upload your SD-generated images
- The tool automatically detects SD metadata patterns
- Click Process to remove all AI signatures
- Download clean images ready for sharing
The tool removes PNG chunks, EXIF data, and embedded workflow information while preserving image quality.
Method 2: A1111 Settings
In Automatic1111, you can disable metadata embedding:
- Go to Settings > Saving images
- Uncheck "Add generation parameters to PNG info"
- Uncheck "Save prompt to file when saving image"
Note: This only affects new images. Previously generated images still contain metadata.
Method 3: ComfyUI Settings
In ComfyUI:
- Open Settings (gear icon)
- Find "Prompt" section
- Disable "Save metadata to files"
Or use a custom save node that strips metadata before saving.
Method 4: Manual Removal with Tools
ExifTool:
exiftool -all= -overwrite_original image.png
Python PIL:
from PIL import Image
img = Image.open('image.png')
img.save('clean.png', pnginfo=None)
ImageMagick:
convert image.png -strip clean.png
SDXL-Specific Considerations
Stable Diffusion XL images may contain additional metadata:
- Refiner settings: If using the two-stage SDXL process
- Base/refiner switch point: Denoising strength where refiner takes over
- Aspect ratio data: SDXL's flexible resolution handling
- VAE information: Which VAE was used for decoding
Our tool handles all SDXL-specific metadata patterns across all major interfaces.
Why Remove Stable Diffusion Metadata?
Privacy Concerns
Your prompts can reveal:
- Personal interests and preferences
- Business or creative strategies
- Technical skill level
- Which models and LoRAs you use
Platform Detection
Social media platforms increasingly scan for AI generation metadata:
- Pinterest may flag images with SD parameters
- Instagram looks for AI tool signatures
- Stock sites may reject images with generation data
Professional Use
For commercial work:
- Clients may not want AI generation disclosed
- Portfolio images should focus on the art, not the process
- Licensing agreements may require "clean" deliverables
Best Practices for SD Metadata Management
Generate with metadata, clean before sharing: Keep generation data for personal reference, remove before publishing
Use separate workflows: Have a "sharing" workflow that automatically strips metadata
Batch process: Clean multiple images at once using our AI Metadata Cleaner
Verify removal: Always check your images with a metadata viewer before sharing
Keep originals: Store unprocessed versions with metadata for your records
Conclusion
Stable Diffusion's open nature means metadata handling varies by interface, but all versions embed data that can identify your images as AI-generated. Whether you're using A1111, ComfyUI, or any other SD interface, removing this metadata is essential for privacy and avoiding AI detection.
Use our AI Metadata Cleaner to remove Stable Diffusion metadata from all interfaces with one tool. For comparison with other AI generators' metadata, see our DALL-E vs MidJourney vs Stable Diffusion guide.

