AI Models
Tools run small open-source models directly in your browser. A model is downloaded once from a CDN, stored in your browser cache, and reused by every tool that needs it. You can delete cached models at any time — they never contain any of your data.
Xenova/modnet
checking cache…MODNet portrait matting, ONNX port by Xenova. Default model for Remove Background. Proven in official WebGPU demos; strong on people and single-subject products. Self-hosted at /models/Xenova/modnet/.
Task: background-removal · License: Apache-2.0 · Size: ~24.8 MB · Runtime: webgpu / wasm
onnx-community/ormbg-ONNX
checking cache…Open Remove Background Model (IS-Net based). Better on general objects than MODNet. Alternative candidate pending /benchmark results.
Task: background-removal · License: Apache-2.0 · Size: ~45.8 MB · Runtime: webgpu / wasm
briaai/RMBG-1.4
checking cache…Non-commercial licenseBest quality but NON-COMMERCIAL license and gated repo. Forbidden in production per MODEL-LICENSES.md policy. Requires a commercial agreement with BRIA AI.
Task: background-removal · License: bria-rmbg-1.4 (CC-BY-NC, non-commercial only) · Size: ~42.9 MB · Runtime: webgpu / wasm
onnx-community/whisper-tiny
checking cache…Default model for Speech to Text. Subtitle Generator will reuse the same cached download. Self-hosted at /models/onnx-community/whisper-tiny/.
Task: automatic-speech-recognition · License: Apache-2.0 · Size: ~40.1 MB · Runtime: webgpu / wasm
paddlejs/pp-ocrv2-det
checking cache…PaddleOCR PP-OCRv2 text detection, paddlejs fused build. Self-hosted at /models/paddle-ocr/det/. Used by Image to Text and Image to Excel.
Task: ocr · License: Apache-2.0 · Size: ~2.5 MB · Runtime: webgl
paddlejs/pp-ocrv2-rec
checking cache…PaddleOCR PP-OCRv2 recognizer (unified Simplified Chinese + English, dictionary bundled in @paddlejs-models/ocr). Self-hosted at /models/paddle-ocr/rec/. Used by Image to Text and Image to Excel.
Task: ocr · License: Apache-2.0 · Size: ~8.9 MB · Runtime: webgl
bukuroo/yunet-int8
checking cache…YuNet face detection (OpenCV Zoo), int8 ONNX export. Self-hosted at /models/yunet/. Used by Blur Face with a hand-written decode/NMS post-process matching OpenCV's face_detect.cpp.
Task: face-detection · License: Apache-2.0 · Size: ~98 KB · Runtime: wasm
Xenova/swin2sr-classical-sr-x2-64
checking cache…Swin2SR 2x classical super-resolution, quantized ONNX port by Xenova (upstream Swin2SR code is Apache-2.0). Self-hosted at /models/swin2sr/. Used by Image Upscaler via transformers.js.
Task: image-to-image · License: Apache-2.0 · Size: ~20.5 MB · Runtime: wasm
andrelgomes/gtcrn-stream
checking cache…GTCRN streaming speech enhancement (n_fft 512, hop 256, sqrt-Hann, 16 kHz, causal with cache tensors). Self-hosted at /models/gtcrn/. Used by Audio Noise Remover.
Task: speech-enhancement · License: MIT · Size: ~522 KB · Runtime: wasm
License information for every model is documented on the privacy page and in the repository's MODEL-LICENSES.md.