raimove
AI Watermark & Metadata Remover — strips invisible Unicode markers, file metadata and C2PA signatures.
Science →Choose a file
Options
Drop a file here or click to browse
.png · .jpg · .jpeg · .webp · .gif · .bmp · .tiff · .avif · .heic · .svg · .pdf · .docx · .xlsx · .pptx · .odt · .epub · .html · .md · .txt
Result
What gets detected and removed
This list isn't maintained by hand; it comes from the same patterns the cleaner searches for.
- OpenAI (ChatGPT, DALL·E, Sora)
- Anthropic (Claude)
- Google (Gemini, Imagen, SynthID)
- Microsoft Copilot
- Midjourney
- Stable Diffusion
- Adobe Firefly
- xAI (Grok)
- Meta (Llama)
- Samsung (Galaxy AI)
Detected in metadata of images, documents and HTML.
- C2PA / Content Credentials
- IPTC (digitalSourceType, AISystem)
- TC260 AIGC label (GB 45438-2025)
- • Zero-width characters (ZWSP, ZWJ, word joiner)
- • Bidi control characters (LRM, RLO …)
- • Variation selectors
- • Tag characters (U+E0000 block)
- • Space homoglyphs (NBSP, em space …)
- • Image metadata (EXIF, XMP, PNG text chunks)
- • Document metadata (PDF, DOCX, ODT)
- • HTML provenance (meta tags, JSON-LD)
- • Generic “AI-generated” tags
Statistical token watermarks are not in scope here: they live in the word choice itself and can only be addressed by the LLM rewrite (Layer B), without guarantees.
For developers: MCP server
Access openRaimove speaks the Model Context Protocol. An agent — Claude Code, Claude Desktop or a client of your own — can clean source code and text directly, without anyone pushing files through this page.
raimove_clean_textclean text, optional LLM rewriteraimove_inspect_textanalyse only, change nothingraimove_clean_filea whole file, source code includedraimove_coveragewhat gets detected and removed
claude mcp add --transport http raimove https://raimove.com/mcp \
--header "Authorization: Bearer <token>"Request a token
To be straight about it: since anyone can get one, the token is an identifier with an expiry date rather than access control. The web interface above does the same work without one.
Over MCP, files are capped at 1.00 MB — well below this page's limit, because the result travels back through a model's context window. Anything larger belongs in the cleaner above.