SynthID Remover & Detector — Images and Video

SynthID is the watermark most tools can't touch. It isn't metadata — it's a pattern Google DeepMind pioneered that embeds in the frequency layers of the pixels (for images) and across the frames (for video), engineered to survive re-saving, resizing, cropping, screenshotting, recompression, and transcoding. That's exactly why an ordinary EXIF or metadata stripper leaves it fully intact, and why removing it needs pixel- and frame-level processing rather than a container edit.

SynthID is now industry-wide: Google, OpenAI, NVIDIA, Kakao, and ElevenLabs all embed it, so it turns up on files from many different tools. This is the single canonical SynthID hub — detector and remover, image and video, every SynthID source in one place. Below we explain how SynthID actually works, what removal can and can't do, and how to verify your result.

SynthID, pioneered by Google DeepMind, marks AI media by embedding an imperceptible pattern into the frequency layers of the image — the same domain a JPEG compressor works in — or across every frame of a video, rather than writing a tag into the file header. Because the signal lives in the content itself, it persists through the everyday edits that wipe normal metadata: cropping, resizing, format conversion, screenshots, moderate compression, and re-encoding all leave it readable by a SynthID detector.

That resilience is why the real question isn't "is there a SynthID tag to delete" but "can the embedded signal be disrupted." A SynthID remover works by applying controlled, targeted adjustments to the pixel or frame values in those frequency bands to break up the watermark pattern, while keeping the visible media as close to the original as possible. Effectiveness depends on the file — resolution, how heavily it was compressed, and the specific SynthID version all matter — so the honest framing is strong signal disruption, not a guaranteed byte-perfect erase.

SynthID is usually layered with other marks, and the mix depends on the source. Google exports frequently add the visible Gemini sparkle logo painted into a corner or burned into frames; OpenAI pairs SynthID with C2PA Content Credentials and, on some downloads, a visible CR badge; and nearly every provider now includes C2PA metadata. Each is a different layer needing different handling — a metadata stripper misses SynthID, and a SynthID pass alone leaves a visible badge — which is why one dedicated workflow that clears all of them matters.

This tool detects and removes every supported layer together, for both media types: the SynthID signal from the pixels or frames, the visible Gemini sparkle logo where present, and the C2PA and file metadata — across Gemini, Nano Banana, Imagen, Veo, ChatGPT / GPT Image 2, and other SynthID-marked media.

Google media carries two separate marks and they need two different answers. The visible sparkle logo sits in the corner and can be lifted cleanly. SynthID is woven through the pixels themselves, and Google builds it to survive cropping, resizing, recolouring and recompression — which is exactly why no browser filter removes it, whatever it claims.

The pass on this page does the first job: the logo, the C2PA credentials and the embedded metadata, all locally, with nothing uploaded. SynthID removal is a different operation entirely and runs in the full workflow, which produces media carrying no SynthID signal at all rather than a weakened one. If that is what you need, open the SynthID workflow.

SynthID is not one product's watermark. It runs across Google's image and video models, so the same signal is in a Gemini picture, an Imagen render and a Veo clip, embedded the same way each time.

For the image side with the sparkle logo included, the Gemini page and the Imagen page are the brand-specific versions. For video, where the signal sits across frames rather than in a single raster, the Veo page and the Gemini video page cover it.

Everything on this page concerns images, where the mark lives in metadata or in the pixels themselves. Text carries a different family entirely: invisible Unicode characters sitting between the words, which render as nothing and survive every copy and paste.

If that is what you are chasing, the ChatGPT watermark remover reports every hidden character it finds by code point, with versions for Claude, Gemini and Grok. The ChatGPT text cleaner on the homepage runs the whole set of passes at once.