A developer has released pagedMark, a tool that removes AI watermarks and provenance metadata from self-generated images and video.
Unlike simple screenshots, it targets invisible pixel-embedded marks (such as SynthID watermarks) by regenerating the image, though this process may alter faces, text, and fine details.
The tool currently supports watermark removal from ChatGPT, gpt-image API, Z-Image Turbo, Nano Banana, and several video generators including Sora and Veo.
What happened
A developer created pagedMark, a tool designed to remove AI provenance markers—both visible labels and invisible pixel-embedded watermarks—from images and videos generated by AI tools. The tool regenerates images to eliminate invisible marks while attempting to preserve visual similarity, and currently supports watermarks from ChatGPT, gpt-image API, Z-Image Turbo, and Nano Banana for images, plus visible marks and metadata from Sora, Veo, Seedance, Hailuo, and Kling for video.
Why it matters
AI-generated content increasingly carries metadata and embedded signals (like SynthID watermarks) that identify its origin. For creators working with their own AI-generated material, these marks can complicate workflow—simple removal methods like screenshots don't eliminate pixel-level markers. pagedMark addresses this by regenerating affected content, though users should know the output may differ from the original (faces, text, and small details can shift).
What to watch
The tool has been tested on Apple Silicon Macs with 8 GB and 16 GB of memory, with memory-aware processing built in to handle resource constraints on lower-spec machines.
Ask the AI about this article →
AI-generated content now routinely carries dual identification signals: human-readable metadata (EXIF, XMP, IPTC tags) and machine-embedded watermarks invisible to the eye, such as SynthID style marks inserted at the pixel level during generation. For creators working with their own AI-generated material, these marks can create friction—they persist through normal file operations and screenshots, making them difficult to remove without specialized tools.
pagedMark tackles this problem by addressing both layers of provenance. The metadata removal is straightforward, but the invisible marks require image regeneration—a computationally heavier approach that trades fidelity (some visual details will shift) for provenance removal. This trade-off is explicit in the tool's design: the developer notes that faces, text, and small details may change, with the goal of minimizing visible degradation while eliminating the watermark signal entirely.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Coca-Cola faced consumer backlash in late 2024 for releasing fully AI-generated holiday ads designed to echo i…
Adobe is releasing three AI audio tools — Generate Music (royalty-free music for videos), Generate Speech (scr…
Researchers at MIT's Computer Science and Artificial Intelligence Laboratory found that generative AI models t…
Higgsfield AI, a startup, produced The Cully Hill Boys, a 110-minute action-comedy featuring licensed celebrit…
A study examining AI-generated images found that many cannot be reliably traced back to their original trainin…
A post on LessWrong poses the question of whether a large language model (LLM)—an AI system that understands a…
