ClearPixOpen-source licenses & model attributions
ClearPix runs entirely on open AI models and open-source software, executed on your device. We are grateful to the researchers and maintainers below.
AI models
MI-GAN
BSD 3-Clause — Picsart Research (ICCV 2023)Image inpainting model that rebuilds pixels under removed watermarks.
Source & license text →Real-ESRGAN (x4plus)
BSD 3-Clause — Xintao Wang et al.Super-resolution model behind the 4× image upscaler.
Source & license text →PP-OCRv5 mobile det
Apache 2.0 — PaddlePaddle / BaiduText-detection model used to auto-locate watermark text.
Source & license text →YOLO11 watermark det (community fine-tune)
AGPL-3.0 — Ultralytics YOLO11 base, community ONNX exportOptional deep-detection tier for faint and tiled watermarks.
This community model rides on YOLO11 (AGPL-3.0). ClearPix loads it on demand, unmodified, as a separate downloadable artifact — not compiled into the app. We are evaluating an enterprise-licensed or self-trained replacement; until then the model is OFF by default and only fetched when you enable Deep detection.
Source & license text →Runtime & framework
ONNX Runtime Web
MIT — MicrosoftExecutes all neural networks in your browser (WebGPU / WASM).
Source & license text →ffmpeg.wasm
MIT — ffmpeg.wasm contributors (FFmpeg itself: LGPL/GPL)In-browser video frame extraction and re-encoding.
Source & license text →Sample images shipped with the site are generated in-house and may be used freely for testing. If you believe an attribution is missing or incorrect, please let us know.