RemoveAILabel
Start cleaning
Local metadata processing · nothing uploaded

Remove the
AI label and metadata from your photos

Platforms read C2PA and XMP — not the picture itself.Strip both, plus the EXIF and IPTC that travel with them.Only the file header is rewritten. Not one pixel changes.

Formats
JPG · PNG · WebP · AVIF · HEIC
Max size
15 MB
Per batch
30 files
Runs on
your browser
Uploads
none
  • No sign-up
  • No logs kept
  • Works offline
Inspect a file first ↗·What it cannot do
Drop images here, or click to choose
JPG, PNG, WebP, AVIF, HEIC · 15 MB each · 30 per batch
You can also paste a screenshot with ⌘V / Ctrl+V
Cleaning options
05
image containers
00
file uploads
30
per batch
100%
on your device

What gets removed

All of it sits in the file header. The cleaned file is pixel-identical to the original — it simply stops carrying the record.

C2PAContent credentialsThe first thing a platform checks before adding an AI label
XMPGeneration parametersWhere prompts, seeds and model IDs land by default
EXIFCamera & locationBody, lens, timestamps, GPS coordinates
IPTCAttributionThe AI attribution fields Adobe tools write
PNG textWorkflow JSONThe full node graph from SD and ComfyUI
ContainersMP4 · PDFudta tags and the Document Info dictionary

Suggested order

Blind cleaning works too. But looking at what is actually inside the file first avoids the usual confusion when a platform still flags it — that normally means the trigger was never in the metadata.

01→
Inspect

See what the file actually contains

02→
Clean

Rewrite the header, leave pixels alone

03→
Verify

Scan the downloaded copy

04→
Upload

Use that copy, edit nothing in between

What each tool leaves behind

Different tools write to different places. That placement decides which block has to go, and why clearing XMP alone is so often not enough.

This is the most common false positive: a real photograph that went through one Generative Fill or AI denoise pass, exported with C2PA attached.

MidjourneyXMP parameters + PNG text chunksDon't stop at XMP
DALL·E / ChatGPTC2PA content credentialsC2PA is the main one
Adobe FireflyC2PA signature + IPTC attributionClear both
Stable DiffusionFull parameters in PNG tEXt / iTXtXMP alone is not enough
ComfyUIWorkflow JSON + node parametersOften hundreds of KB
Photoshop Generative FillLocal C2PA on a real photoThe usual false positive
Lightroom AI DenoiseAI markers in XMP / C2PAUse precise mode
Canva / CapCutSoftware fields and container tagsClean video and cover separately

Common questions

Something more specific? Start withwhat this cannot do— it spells out the cases that are out of reach.

Does cleaning affect image quality?+

No. Metadata and pixels are separate data. Only the header blocks are rewritten; pixel bytes are copied verbatim and never pass through an encoder.

Are my files uploaded anywhere?+

No. Parsing, cleaning and zipping all run in your browser. Once the page has loaded you can cut the network and it keeps working — that is the most direct proof available.

The platform still flags it as AI afterwards. Why?+

The trigger was never in the metadata. It may be an invisible pixel-level watermark such as SynthID, or the platform's own visual classifier. Neither lives in the file header.

What counts as a false positive?+

A real photograph that went through one round of Generative Fill, AI denoise or AI background replacement picks up C2PA. Platforms do not distinguish partial assistance from a fully generated image.

Why keep the EXIF option at all?+

Photographers hand body and lens details to clients. Turn off “remove EXIF” and only the AI markers go; the camera data stays exactly as it was.