What ScrubPhoto cannot do

ScrubPhoto removes metadata: the hidden fields a file carries next to the image, like C2PA Content Credentials, IPTC and XMP fields, EXIF camera data and GPS location. It never touches the pixels, the video frames or the audio. That is why the result is lossless, and it is also why some things are out of reach for this tool and for any other metadata tool.

This page lists every limit we know of. If a platform still labels your post after a scrub, one of these is almost always the reason.

Invisible watermarks live in the pixels

SynthID is an invisible watermark from Google DeepMind. It is written into the pixels of an image, the frames of a video or the samples of an audio track, so removing metadata leaves it in place. Google uses it on images and video from its own models. OpenAI's help center says images generated with ChatGPT, Codex or the OpenAI API carry SynthID as well as Content Credentials. Apple said in June 2026 that photos adjusted with Apple Intelligence will carry a hidden SynthID watermark, and in September that the feature arrives in a software update later this year.

Meta adds its own invisible watermark to images made with Meta AI, alongside IPTC metadata (Meta newsroom, Feb 6 2024). TikTok said in November 2025 that it adds invisible watermarks to content made with its AI tools and to uploads that carry C2PA, and that only TikTok can read them.

ScrubPhoto does not detect any of these watermarks and does not remove them. It will not give tips for defeating them either. If a file came out of one of these tools, assume the watermark is still there after a scrub. Read more in Can you remove SynthID?.

Visible watermarks are part of the image

Some apps draw a mark into the picture itself. Samsung puts a Galaxy AI watermark in the bottom-left corner of images made with its generative editing tools, and Meta AI images can carry an "Imagined with AI" mark. These are pixels, not metadata, so ScrubPhoto leaves them exactly as they are. The Samsung guide covers what Samsung writes.

A label you added yourself stays

Instagram, Facebook, Threads, TikTok and YouTube all let creators add an AI label by hand. That label is stored with the post, not in the file, so nothing about the file can change it. Turn it off before you share if the post does not need it, and check the post's settings if you added it by mistake. The Instagram toggle guide explains the difference between the two kinds of label.

Platforms run their own detection

Metadata is one signal among several. Meta said in February 2024 that it is building classifiers that can spot AI content without markers. TikTok layers its own detection models on top of labels and Content Credentials. YouTube says it can apply labels from its internal detection. A clean file removes the metadata trigger these platforms document. It does not stop a model that looks at the image itself.

A label on a live post stays on that post

Scrubbing works on the file on your device. It cannot reach a post that is already published. TikTok's help center says its automatic AI-generated label cannot be removed once applied, and YouTube says labels that come from C2PA cannot be adjusted. On Instagram and Facebook the honest options are to keep the post, or to delete it and post a scrubbed copy, which loses the likes and comments. The after-posting guide walks through it.

Disclosure rules still apply

Removing metadata does not change what a platform asks of you. Meta requires people to disclose photorealistic video and realistic audio that was made or changed with AI, and may apply penalties when they do not. TikTok requires creators to label realistic AI content, and YouTube asks for disclosure of realistic AI changes. Some AI tools also forbid removing provenance tags in their own terms. Canva's AI Product Terms, effective June 26 2026, prohibit removing provenance or metadata tags, including C2PA, from AI-generated content, and Google's generative AI use policy forbids misrepresenting where generated content came from. Use ScrubPhoto on files you have the right to change, for the reasons it was built: privacy, and fixing labels on real photos.

You can lose a good label too

Content Credentials are not only about AI. Instagram has a Captured with a camera label that relies on secure capture data in C2PA, and YouTube has a similar camera label. Pixel 10 phones and some Leica, Nikon and Sony cameras sign their photos this way. Scrub a camera-signed file and it can no longer prove where it came from. If proof of capture matters to you, keep the original.

Format limits

• HEIC, HEIF and AVIF are not lossless in this version. Where your browser can open them (Safari for HEIC), ScrubPhoto re-saves the photo as a JPEG at quality 95 and says so. Chrome and Android cannot open HEIC, so convert to JPEG in your Photos app first.
• Some videos carry a timed metadata track inside the video stream. ScrubPhoto leaves those tracks in place because removing them would mean rebuilding the file. The checker shows when a video has one.
• PDF, Word files, audio-only files and RAW camera files are not supported.
• A damaged file is refused with a message instead of being guessed at.

Questions about the limits

Will a scrubbed photo never get an AI label?

No tool can promise that. A scrub removes the metadata signals that Meta and TikTok document, which is the most common trigger. Watermarks, self-added labels and platform detection can still lead to a label.

Does ScrubPhoto detect AI images?

No. It reads what a file says about itself in its metadata. It cannot tell whether the pixels were made by AI, and it never claims to.

Why does ScrubPhoto not offer to add camera data?

Writing a camera make or model into an AI image invents a history the file never had. Removing data and making up data are different acts, and ScrubPhoto only does the first.