Upload a supported image
Sign in and add your ChatGPT or DALL·E image in JPEG, PNG, WebP, AVIF, HEIC, or HEIF, up to 20 MB.
ChatGPT watermark remover
ChatGPT & DALL·E Watermark Remover
Images from ChatGPT and DALL·E can carry C2PA Content Credentials. If a file also contains a SynthID pixel layer, that is a separate signal. We strip provenance metadata and regenerate pixels remotely. See removing C2PA Content Credentials and removing SynthID.
First 3 images freeNo subscriptionRefund on failed jobs
How it works
Upload. We clear the credentials. You download.
Content Credentials and the embedded signal, in one private pass.
We process it privately
We strip the C2PA manifest and “Made with AI” metadata, then regenerate pixels in an isolated, no-network container to disrupt the embedded SynthID layer.
Download the result
Retrieve the finished image from your signed-in account.
What it removes
What OpenAI images carry today.
ChatGPT and DALL·E images are labeled in more than one way. Clearing the visible “Made with AI” note is not the same as clearing the signed manifest or the pixel signal.
C2PA Content Credentials
A signed C2PA manifest (JUMBF/APP11 in JPEG, caBX in PNG) plus “Made with AI” XMP/EXIF fields are stripped from the file.
SynthID pixel layer
A SynthID pixel layer is separate from the C2PA manifest, so stripping metadata leaves it intact. Pixel regeneration attempts to disrupt it; removal is not guaranteed.
Any visible label
Where a visible “Made with AI” bar or badge is present, the region is reconstructed so it blends into the surrounding image.
ChatGPT and DALL·E provenance
How OpenAI marks images, layer by layer.
C2PA Content Credentials
OpenAI attaches C2PA Content Credentials to ChatGPT and DALL·E images: a cryptographically signed manifest embedded as a JUMBF/APP11 segment in JPEG or a caBX chunk in PNG, alongside “Made with AI” XMP and EXIF fields. Anyone can read it at contentcredentials.org/verify. We remove this in the same way described on the C2PA removal page and in removing C2PA Content Credentials.
A pixel layer, when present
An image can carry a SynthID pixel-level watermark as well as a C2PA manifest. Unlike the manifest, this signal lives in the pixels and is not cleared by stripping metadata. It requires pixel regeneration, and results must be verified. Read what SynthID is to understand why.
Why a metadata wipe is not enough
Deleting Content Credentials removes the signed provenance record, but the embedded SynthID signal can remain and still be detectable. A complete pass has to handle both, the same reasoning behind our general AI watermark remover. Verify results at contentcredentials.org/verify and with Google’s SynthID Detector.
Formats and private processing
Uploads can be JPEG, PNG, WebP, AVIF, HEIC, or HEIF up to 20 MB. The metadata primer explains where these labels live, and private, offline processing covers how your image is handled in isolation.
Straight answers
ChatGPT and DALL·E watermarks, answered.
Do ChatGPT and DALL·E images have a watermark?
They can carry C2PA Content Credentials, including a signed manifest and AI provenance metadata. A file may also contain a pixel-level signal; inspect and verify the actual file rather than assuming every output has the same layers.
How do I remove the ChatGPT watermark?
Upload the image and we strip the C2PA manifest and metadata, reconstruct any visible label, and regenerate pixels to disrupt the embedded SynthID layer. The first 3 images are free.
Is there a visible watermark on DALL·E images?
DALL·E images previously used a colored swatch in the corner and now rely mainly on C2PA Content Credentials. Where a visible label exists, we reconstruct that region; the invisible credentials and signal are handled separately.
Can I just delete the metadata myself?
You can strip Content Credentials, but that leaves the embedded SynthID pixel layer intact and still detectable. Removing that layer requires regeneration, not a metadata wipe.
Is this affiliated with OpenAI or Google?
No. Delete SynthID is an independent service, not affiliated with, endorsed by, or operated by OpenAI or Google.
Related tools
Choose the workflow that matches your file.
For images you own.
Use this only on ChatGPT or DALL·E images you generated or have the right to edit. Results vary and no tool guarantees every credential or embedded signal is removed. Delete SynthID is independent and not affiliated with or endorsed by OpenAI or Google. See responsible editing.