Remove ChatGPT metadata from images: what goes and what stays.
Images from ChatGPT, Codex and the OpenAI API carry C2PA Content Credentials and a SynthID watermark. This tool removes the C2PA metadata in your browser. It cannot remove the watermark, and here is why that matters.
What OpenAI puts in its images
OpenAI states that supported images generated with ChatGPT, Codex and the OpenAI API "include both signals": C2PA metadata and SynthID watermarks (OpenAI Help Center). The two are very different:
| C2PA metadata | SynthID watermark | |
|---|---|---|
| What it is | A signed record attached to the file: the tool that created it, when, and other details about its origin | An invisible pattern in the pixels |
| Since when | 2024, starting with DALL·E 3, later ImageGen and Sora (OpenAI) | Announced in 2026 through a partnership with Google (OpenAI) |
| Removed by this tool | Yes | No |
OpenAI is open about the limits of metadata. It says C2PA "can be stripped, lost through uploads and downloads, or broken by transformations like file format changes, resizing, or screenshots" (OpenAI), which is exactly why it added a watermark that lives in the image itself. It also notes that coverage "can vary by product, model, export path, file type, and when the content was created" (OpenAI Help Center), so not every file looks the same. Checking your own file is the only way to know what it carries.
Check and remove the metadata
- Drop the image you downloaded from ChatGPT or the API. It is read in your browser, never uploaded.
- The report shows whether a C2PA manifest is present, whether it marks the content as AI-generated, and which tool or company names it mentions.
- Download the cleaned copy. The pixels are identical to the original, because the image data is copied byte for byte.
A screenshot also drops the metadata, as OpenAI itself notes, but it re-encodes the picture, can change its size and adds your device's own metadata. Removing the manifest directly keeps the original image intact.
What still identifies the image
After cleaning, the SynthID watermark is still in the pixels. OpenAI runs a checker at openai.com/verify for supported images and audio (OpenAI Help Center), and Google's Gemini app can check for SynthID too. Whether a social platform looks for the watermark is that platform's decision. In practice:
- Metadata removal clears the signals that platforms like Meta and TikTok describe reading: C2PA and IPTC.
- It does not make an image untraceable, and it does not change what the image is.
More detail on the watermark is on the SynthID page.
Sora videos and OpenAI audio
The same approach covers OpenAI's video and audio. OpenAI says it added Content Credentials to Sora as well as to images (OpenAI), and in July 2026 it added SynthID watermarking to supported audio generated with its tools, including through ChatGPT and the API (OpenAI). For a Sora MP4, the C2PA manifest is a box in the file, which the video metadata remover blanks without re-encoding. A watermark carried in the picture or the sound is a different matter, and it stays.
Good reasons to clean a ChatGPT image
- You use generated images as drafts, mockups or illustrations in your own work, disclose AI use where it matters, and don't want a provenance record about your account and workflow travelling with every copy.
- You combine a generated element into a larger design and prefer to describe the AI use in your own words.
- You want to strip all metadata from files before handing them to a client or publishing them on your own site.
Frequently asked questions
Does taking a screenshot remove ChatGPT metadata?
Usually, and OpenAI lists screenshots among the ways metadata gets lost. But a screenshot re-encodes the image and adds your device's own metadata, and it does not remove the SynthID watermark.
Do Sora videos have metadata too?
OpenAI says it adds Content Credentials to Sora videos. The video metadata remover clears the C2PA box from MP4 files without re-encoding.
Can I turn off C2PA in the OpenAI API?
We have not found an option for that in OpenAI's documentation. Removing the metadata from the downloaded file is the way to get a copy without it.
Do ChatGPT images have metadata?
Yes. OpenAI says supported images from ChatGPT, Codex and the OpenAI API include C2PA metadata and a SynthID watermark.
Can this remove the SynthID watermark from ChatGPT images?
No. SynthID is embedded in the pixels. This tool only removes metadata and never changes the image data.
Does removing the metadata change the image?
No. The image data is copied byte for byte, so the cleaned file shows exactly the same pixels.
How can I check whether my image still has provenance signals?
Drop the cleaned copy into our AI metadata checker for metadata. For the watermark, OpenAI offers openai.com/verify.