Photo SignalsAI Watermark Remover

Sora watermark remover — for stills and exported frames, not the video

OpenAI closed the Sora app on 26 April 2026 and told everyone to export their videos. The files people kept still carry everything Sora wrote: a visible watermark in the picture, a C2PA record in the container, and a fingerprint that is not in the file at all. This tool opens an image — a frame you exported, or a screenshot — and says plainly what it does not reach.

  • The visible watermark
  • C2PA metadata
  • OpenAI's own search

Open an image and paint over the mark

  • Nothing is uploaded
  • No account
  • Free, no limit
What Sora actually wrote

Three signals, and only one of them is paint

  • 1 · The visible watermark

    Removed from a frame, and you can see it is gone

    OpenAI's own words: every video generated with Sora carries a visible watermark, on by default. In a video it is drawn into the frames; in a frame you export it is a static badge, which is paint on the picture — so you paint over it here and it fills in from the surrounding image. OpenAI's pages do not say where the mark sits, and people who exported their clips describe a logo that changes position through the video. Either way, a mask that is right on one frame is wrong on the next, which is one reason this tool takes an image and not a video.

  • 2 · C2PA metadata

    Removed from an image with proof; not reached inside the MP4

    Every Sora video embeds a signed C2PA record naming its origin. That record sits in the video container, and this tool does not open video containers — it will not pretend to. A frame you export to PNG or JPEG carries whatever the exporting program wrote: if that is a C2PA record, generator fields or an editing history, it is listed, dropped in the rebuild, and the result is read back with the same parser to prove it is gone.

  • 3 · OpenAI's own search

    Not in your file — so not claimed

    OpenAI keeps internal reverse-image and audio search tools that it says trace videos back to Sora with high accuracy. That is a provider's own record of what it generated. Nothing about it is stored in your file, so nothing done to the file can change it, and nobody outside OpenAI can test a frame against it. The findings list names this class "not verifiable" rather than leaving it off.

If you still need the video itself. Sora's export window is at sora.chatgpt.com/sunset while OpenAI keeps it open; the API follows on 24 September 2026. Nothing here processes an MP4, and a page that says otherwise about a moving mark is describing a re-render, not a removal.

Four steps

Export a frame, paint the badge, clean the container

  1. Get a still

    Export the frame you want from any video player, or take a screenshot. That is the file this tool works on; the video stays as it is.

  2. Paint over the badge, or let the model suggest it

    Suggest the mark for me proposes a mask; nothing is removed until you have edited it and pressed Fill. The filled result is written as a fresh PNG, so it carries no record from the original.

  3. Remove whatever the exporter wrote, with proof

    Press Remove metadata, C2PA and hidden files instead on the same file. Container only leaves every pixel byte-identical; the deeper levels re-encode and measure what that moved.

  4. Read the numbers before you trust the file

    The result is re-inspected and the page reports what was removed, what was reduced by how much, and what it cannot verify.

What is measured afterwards

Every result is read back by the same engine that produced it. These are the readings you will see, and what each one is allowed to mean.

  • Checked again on the result

    The container, re-read

    The cleaned file is parsed a second time. If a second pass would still change bytes, the page says so instead of calling the file clean. This is the proof that a manifest or a generator field is gone from the frame.

  • Perceptual fingerprint moved N of 64 bits

    At the re-encode level

    How far the re-encode moved the picture's perceptual hash. Usually very little, and the page says so: nudging pixels does not defeat a provider's lookup, and this number is there so you are not told otherwise.

  • High-frequency band reduced by X dB

    At the deepest level

    What the learned pass did to your image, as a measurement. On our benchmark the model reduces an injected residual by 8.5 dB while keeping the picture at 38.9 dB PSNR — a synthetic residual, not any mark of Sora's, and the page labels it that way.

What this page will not say. It will not say it removed a Sora watermark from a video, because it never opened one. It will not tell you OpenAI can no longer recognise the frame, because that search is not in your file and nobody outside can test it. What it removes from an image, it proves by reading its own output back; what it only reduces, it reports as a number.

Sources: OpenAI, "Launching Sora responsibly", OpenAI, "Sora is here", OpenAI on the Sora discontinuation. Other generators: Gemini and Nano Banana, every AI mark, the limits.