ComfyUI-remove-ai-watermarks

ComfyUI custom nodes for remove-ai-watermarks: remove visible AI watermarks, erase arbitrary regions, detect known marks, and strip invisible SynthID watermarks by diffusion regeneration, all inside a ComfyUI graph.

Nodes

Node	What it does
Remove Visible Watermark (RAIW)	Removes a known visible mark: Gemini / Nano Banana sparkle, Doubao "豆包AI生成", Jimeng "★ 即梦AI", Samsung Galaxy AI. mark = auto picks the strongest detected one. cv2-only, no GPU.
Detect Visible Watermark (RAIW)	Reports per-mark detection confidence on the input image. Outputs a text report, a detected boolean, the best confidence, and the detected mark key.
Erase Region (RAIW)	Inpaints whatever a MASK covers. cv2 backend (fast, no deps) or lama backend (big-LaMa via onnxruntime, better quality).
Remove Invisible Watermark / SynthID (RAIW)	SDXL diffusion regeneration that defeats the SynthID pixel watermark while preserving text/face structure (canny ControlNet). Requires the GPU/ML extra.

All pixel nodes operate in-memory on the ComfyUI image batch. The invisible node writes each frame to a temp file, runs the diffusion engine, and reads it back.

Install

Via ComfyUI Manager

Search for "Remove AI Watermarks" in ComfyUI Manager and install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/wiltodelta/ComfyUI-remove-ai-watermarks
pip install -r ComfyUI-remove-ai-watermarks/requirements.txt

The base install (remove-ai-watermarks) covers the three pixel nodes. The invisible / SynthID node additionally needs the diffusion stack (torch, diffusers; multi-GB):

pip install "remove-ai-watermarks[gpu]"

Optional backends: pip install "remove-ai-watermarks[lama]" for the LaMa erase backend, pip install "remove-ai-watermarks[esrgan]" for the Real-ESRGAN upscaler in the invisible node.

Notes

• ComfyUI image tensors carry no file metadata, so the invisible node's vendor-adaptive strength default falls back to the unknown-vendor value (0.30). Set the strength input above 0 to override it.
• There is no local SynthID detector; verify removal with the Gemini app's "Verify with SynthID" oracle. Higher strength removes more but drifts further from the original.
• remove-ai-watermarks depends on opencv-python-headless. If your ComfyUI install already ships opencv-python, both provide cv2 and coexist; if you hit a cv2 conflict, keep a single OpenCV distribution in the environment.

License

Apache-2.0, matching the upstream library.