Remove Stable Diffusion & ComfyUI PNG Metadata
Drop images here or click to upload
JPG, PNG, WebP, AVIF, HEIC · Max 15MB · Batch up to 30 · Not GIF
ComfyUI graphs and A1111 seeds live inside PNG text chunks — not just EXIF.
Recommended workflow: metadata checker → remover
Each AI exporter leaves different fingerprints in your file. Inspect read-only before you strip fields — then use the remover on this page for the upload you will publish.
- Step 1 — inspect your file: AI metadata checker.
- Upload your export to the AI metadata checker (read-only, browser-only)
- Review C2PA, XMP, PNG chunks, and EXIF software strings
- Use the remover above with matching options enabled
- Download the cleaned JPG, PNG, or WebP
- Upload to Instagram, Facebook, Pinterest, TikTok, or LinkedIn
After Photoshop, Canva, or compositing
A raw AI download is not automatically upload-ready. Generative Fill, Canva AI backgrounds, or compositing can add C2PA or IPTC AI fields on top of the original export. Always clean the final flat raster you publish — not intermediate layers or PSD masters.
Why SD / ComfyUI PNGs trigger AI labels
Unlike a camera RAW pipeline, diffusion exports often carry full workflow serialization inside the PNG container: node graphs, prompts, seeds, and checkpoint names.
That is valuable for reproducibility — and it is also a loud automated signal for "this file came from a generator". Inspectors and community tooling have long surfaced PNG chunk data on creative uploads.
A1111 (SD Web UI) vs ComfyUI vs headless inference
- SD Web UI (A1111) — classic parameter + PNG chunk patterns; watch embedded PNG info when "Save metadata"-style options are on.
- ComfyUI — workflow JSON in PNG text chunks is the default mental model.
- API / batch workers — confirm your writer does not reintroduce chunks after post-processing.
Clean this file before you upload
Strip C2PA and XMP in your browser, then publish the cleaned copy.
How this differs from Midjourney and DALL·E
Each toolchain emphasizes different metadata classes:
- Midjourney leans on XMP.
- DALL·E leans on C2PA.
Stable Diffusion front-ends lean on PNG ancillary text. The remover above hits all three classes so you do not need separate desktop apps per toolchain.
Recommended cleaning checklist
- Flatten to the final raster (PNG or JPG) you will ship.
- Enable PNG text chunk removal for PNGs from SD stacks.
- Keep XMP enabled if you passed through another editor afterward.
- Keep C2PA enabled if you mixed in Firefly or ChatGPT image assets.
- Confirm with the metadata checker.
Deep dive on the blog
- Read stripping Stable Diffusion metadata for chunk-level detail and batch ideas.
ComfyUI shortcut
Looking for ComfyUI-only wording? We keep a /comfyui URL that redirects here so bookmarks and nav stay stable.
What this tool removes (SD / ComfyUI)
- PNG tEXt / iTXt workflow and parameters chunks
- XMP generation blocks when present
- C2PA when mixed from other tools
- EXIF software strings
Does not remove pixel watermarks. See disclaimer .
FAQ
Where does ComfyUI hide metadata in PNG files?
ComfyUI and many SD Web UI (A1111) builds store serialized workflow graphs inside PNG text chunks (for example parameters or workflow JSON). Those chunks can survive re-encode steps in some tools, so explicitly strip PNG text metadata before social upload.
Does JPG export avoid the problem?
JPG cannot store the same PNG tEXt workflow payload, but exporters may still attach XMP or EXIF software fields. Always clean the exact file you upload.
Is removing PNG chunks lossless for pixels?
Yes. Removing ancillary text chunks does not alter decoded pixel values.
I saved as JPG from Comfy — am I done?
JPG avoids PNG workflow chunks but may still carry XMP/EXIF. Run the final JPG through the same pipeline.
Does latent upscaling remove chunks?
Not reliably. Some nodes re-encode PNG and drop chunks; others preserve them. Never assume — verify with the checker.
Try it now
Confirm PNG chunks are gone with our AI metadata checker.