Strip Stable Diffusion & ComfyUI Workflow Metadata
Stable Diffusion front-ends write the full generation recipe into the image: prompt, negative prompt, seed, sampler, model hash and, for ComfyUI, the entire node graph. Strip it all here.
Drop images here
JPEG, PNG or WebP · up to 25 MB each · add as many as you like
Binary telemetry
- [1/4] ReadFile → ArrayBuffer
- [2/4] ParseWalk container segments
- [3/4] StripDrop selected metadata
- [4/4] RebuiltEmit cleaned bytes
Cleaning profile
Only the categories you check below are stripped.
Write a fresh Make, Model and ISO block into every cleaned image.
No images queued
Add JPEG, PNG or WebP images above to build a batch. You can queue 30+ files and clean them all at once.
Single-image inspector
nothing selectedNo image loaded
Add an image above to inspect its metadata in detail.
Awaiting an image
Drop a JPEG, PNG or WebP above to read its metadata. Parsed fields appear here as a grouped evidence table.
0 B uploaded · 0 bytes sent
How It Works
- 1
Drop a generated PNG or WebP; the parser scans PNG text chunks and XMP for generation fields.
- 2
The inspector surfaces the parameters string, workflow JSON and model hashes it finds.
- 3
AI metadata is pre-selected so every generation field is removed.
- 4
Clean and download an image with no prompt, seed or workflow graph attached.
Binary Format Breakdown
| Segment | Carries | Action |
|---|---|---|
| PNG tEXt (parameters) | Automatic1111 prompt, negative prompt, steps, seed, sampler | Whole text chunk dropped |
| PNG tEXt (workflow / prompt) | ComfyUI node graph and API prompt JSON | Whole text chunks dropped |
| XMP | Model name, hash and generation parameters | Rewritten in place to drop AI fields |
Frequently Asked Questions
What does ComfyUI store in the image?
ComfyUI writes its complete node graph and the API-format prompt as JSON in PNG text chunks, which can reveal every model, LoRA and setting used.
What does Automatic1111 store?
It writes a single parameters string containing the prompt, negative prompt, steps, sampler, CFG scale, seed and model hash.
Will the image still open after cleaning?
Yes. Only the text chunks are removed; the compressed pixel data and the iCCP color profile are preserved.