Quick entry: This page is a focused entry to AI Image Inspector Stable Diffusion / A1111 / ComfyUI. Go to AI Image Inspector→
Drop a PNG, JPEG, or WebP image
AI generation fingerprints are detected in your browser. Nothing is uploaded.
Remove Stable Diffusion Metadata — Strip SD / A1111 / ComfyUI Metadata
Intended use
- The strip tool removes metadata chunks but does NOT invert the CompVis dwtDct invisible watermark. Use the Phase 2 dwtDct Watermark Remover for that.
- Stripping the 'parameters' chunk is irreversible — your prompt, model hash, and seed cannot be recovered from the stripped file. Keep a backup of the original PNG if you might want to reproduce the generation later.
How to use
Drop the PNG you want to clean
Drag a PNG file from your desktop, your browser, an output folder, or the /outputs directory of your Stable Diffusion install into the drop zone. The image is read into a Uint8Array in your browser. Nothing is uploaded — the strip runs as pure JavaScript in the same tab.
Read the Stable Diffusion metadata report
The inspector scans for the three Stable Diffusion-specific text chunks: 'parameters' (the full generation config — prompt, negative prompt, sampler, steps, CFG scale, model hash, seed), 'prompt' (older A1111 format), and 'workflow' (ComfyUI's JSON workflow). It also checks for the CompVis dwtDct invisible watermark. Each detection is shown as a card with the byte-level payload and a one-click strip button.
Strip the metadata and verify
Click strip on each card to remove the corresponding text chunk from the PNG. The inspector rewrites the PNG in place — no re-encode, no quality loss. Re-run the cleaned file through the inspector to confirm the 'parameters', 'prompt', and 'workflow' chunks are gone and the dwtDct watermark no longer scores above the detection threshold.
Frequently asked
How do I remove Stable Diffusion metadata from a PNG?
Drop the PNG into the AI Image Inspector on this page. The inspector scans for the 'parameters', 'prompt', and 'workflow' PNG text chunks that Automatic1111, Forge, SD.Next, and ComfyUI write, and the CompVis dwtDct invisible watermark. Each card has a one-click strip button that rewrites the PNG without the metadata. The whole strip runs in your browser — no upload, no server, no account.
What metadata does Stable Diffusion embed in PNGs?
Stable Diffusion frontends write one or more PNG tEXt chunks. Automatic1111 and Forge write a 'parameters' chunk with the full generation config (prompt, negative prompt, sampler, steps, CFG scale, model hash, seed, size). ComfyUI adds a 'workflow' JSON chunk describing the node graph. Older A1111 builds split the prompt and negative prompt into separate 'prompt' and 'Negative prompt' chunks. Civitai and some extensions also write EXIF tags.
What is the CompVis dwtDct watermark?
Stable Diffusion versions 1.x and 2.x carry an invisible watermark implemented by CompVis using a DWT-DCT steganography scheme. The watermark survives JPEG re-encoding, resizing, and color adjustments under a threshold. It can only be removed by a targeted tool (Phase 2 dwtDct remover), not by a generic metadata strip. Newer Stable Diffusion models (SDXL and later) drop the watermark by default.
Will this work for Automatic1111, Forge, SD.Next, and ComfyUI?
Yes. The inspector detects all four frontends by reading the union of their text chunks. 'parameters' is the A1111/Forge/SDNext canonical chunk; 'workflow' is the ComfyUI canonical chunk; 'prompt' is the older A1111 split format. The strip button removes whichever chunks are present in the file in a single pass.
Will the image quality change after stripping?
No. Stripping the text chunks is a metadata-only operation — the pixel data, the color profile, the PNG filter strategy, and the compression level all stay identical. The output PNG decodes to the exact same image, but with a smaller file size (often 1–10 KB smaller) and no embedded prompt or workflow.
Why should I remove Stable Diffusion metadata?
Three reasons. Privacy: the 'parameters' chunk contains your full prompt, your model hash, and your seed — anyone who downloads your PNG can read them and reproduce your generation. Cleanup: many galleries and stock-photo sites will reject images with embedded SD metadata. Compliance: some content-licensing workflows require the metadata to be stripped before publication.
Are my images uploaded to a server?
No. The image bytes are read into a Uint8Array in your browser. The PNG chunk reader, the strip routine, and every other operation all run as pure functions in JavaScript. No network call, no server roundtrip, no third-party API, no telemetry, no account. The file never leaves your device, and nothing is stored between sessions. You can verify this by watching the Network tab in DevTools while you drop a file in.
Limitations
- PNG only for the chunk stripThe 'parameters', 'prompt', and 'workflow' chunks are PNG-specific text chunks. If your SD output has been exported as JPEG or WebP, those chunks are not present in the output (JPEG and WebP have their own metadata sections). Run the SD export in PNG format to get the metadata-removal benefit, or use the Strip Image Metadata tool for general-purpose metadata wiping.
- dwtDct watermark needs a dedicated tool (Phase 2)The CompVis dwtDct invisible watermark is a pixel-level watermark, not a metadata chunk. Stripping the PNG text chunks does NOT remove the dwtDct signal. A targeted Phase 2 tool (the dwtDct Watermark Remover) is needed to invert the watermark. This page routes to the Inspector for now — the targeted remover ships in Phase 2.
- Screenshots and re-encodes are already strippedIf you took a screenshot of your SD output, or you exported the image as JPEG and re-encoded it, the PNG text chunks are already gone. The inspector will report 'No AI generation traces found' for the chunk detectors. There is nothing left to strip — the metadata was lost at the re-encode step, not the strip step.
Platform notes
- macOS
- Preview.app does not surface PNG text chunks. Quick Look shows the rendered image but not the embedded 'parameters' or 'workflow' JSON. The inspector is the right tool for diagnosing what is inside an SD PNG and stripping it. To see the metadata in Finder, right-click → Get Info → More Info shows EXIF but not PNG tEXt.
- Windows
- Windows Photos and File Explorer do not show PNG text chunks. Right-click → Properties → Details shows EXIF (if any was embedded) but not the 'parameters' or 'workflow' chunks. For full SD provenance, drop the file into the inspector — the report covers every text chunk in one pass and strips them in place.
- Linux
- CLI alternative for the chunk strip: `pngcheck -v image.png` lists every tEXt chunk including 'parameters' and 'workflow'. `exiftool -api RequestAll=3 -overwrite_original -all= image.png` strips every metadata block including PNG tEXt. The inspector is the right pick when you want a one-click per-chunk report instead of a wholesale wipe, or when you need to keep some tags and drop others.
- Web
- Runs entirely client-side. The image is read as a Uint8Array, the PNG chunk reader and strip routine run as pure functions, and the report is rendered as HTML. No upload, no server roundtrip, no telemetry, no third-party calls. The file never leaves your device.