AI watermark remover

Strip C2PA Content Credentials, EXIF, XMP and hidden Unicode from files you own. Nothing is stored.

Drag a file here or click to choose

Processed in memory on our server and discarded right after the response.

Keep non-AI metadata (camera EXIF, document author)

Check	Before	After
C2PA	—	—
AI metadata	—	—
Hidden characters	—	—
Size	—	—

Rewrite your text with a language model and compare the result with your original. The model is instructed to preserve the language, facts, numbers and names; check the result before using it. We do not guarantee watermark removal or any AI detector score.

Rewrite text

Paste the text to rewrite

0 words · Costs 10 credit packages per 1,000 words

What it removes

C2PA manifests

C2PA / Content Credentials manifests (the 'made with AI' label from ChatGPT, Gemini, Firefly and camera apps)

Metadata

EXIF, XMP, IPTC and document properties (software, author, prompt text, timestamps, GPS)

Invisible Unicode

Invisible Unicode in text: zero-width characters, bidi controls, tag characters, exotic spaces

What it does not do

It does not remove pixel watermarks such as Google SynthID. Rewriting text is a separate, paid step below; it changes wording and is never a guarantee against detectors. It does not touch the visible content of your file. If a file still carries a mark we cannot remove, the report says so.

Supported files

Group	Formats
Images	png jpg webp avif heic bmp gif tiff svg
Documents	pdf docx xlsx pptx epub odt
Text	txt md html
Audio & video	mp4 mov m4a m4v wav mp3 flac

Pricing

Free: 5 files a day, up to 25 MB, no account. With a free Ulazai account: 5 credits per file (about 5 cents), up to 100 MB, plus API access. Text rewrite: 10 credits per 1,000 words (about 10 cents), account required.

API reference

API

Clean one file with your API key and get the bytes back:

curl -X POST https://ulazai.com/api/v1/provenance/clean/ \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Accept: application/octet-stream" \
  -F "[email protected]" \
  -o photo-clean.png

Without the Accept header you get the full report as JSON:

{
  "ok": true,
  "kind": "image",
  "filename": "photo-clean.png",
  "content_type": "image/png",
  "before": { "format": "png", "has_c2pa": true, "has_ai_metadata": false, "findings": [] },
  "after": { "format": "png", "actions": ["strip_c2pa"], "bytes_in": 1048576,
             "bytes_out": 1032192, "changed": true, "still_has_c2pa": false,
             "still_has_ai_metadata": false, "post_findings": [] },
  "actions": ["strip_c2pa"],
  "bytes_in": 1048576,
  "bytes_out": 1032192,
  "changed": true,
  "download": { "filename": "photo-clean.png", "base64": "..." },
  "quota": { "remaining": 4 },
  "charged": 0
}

FAQ

Is my file uploaded and kept?

It is processed in memory and discarded right after the response. We keep no copies and no logs of the content.

Can I remove SynthID?

No. SynthID sits in the pixels and survives re-encoding. Use a different image model if you need to avoid it.

Does it change image quality?

No. Metadata blocks are dropped; pixels are copied as-is.

Is this legal?

Use it on content you own or are allowed to edit. Removing provenance from someone else's work can be unlawful.

Which model rewrites my text?

A language model served through OpenRouter. Your text is sent to the model provider for rewriting. We instruct it to keep language, facts, numbers and names unchanged; check the result before using it.