Credits & model licences
Everything below runs on your device. The models are downloaded to your browser and never see a server of ours, so there is nothing for us to receive.
PP-OCRv5 mobile detection head
ppocrv5-det · 4.83 MB · Apache-2.0
Real-ESRGAN realesr-general-x4v3 (SRVGGNetCompact)
realesr-general-x4v3 · 5.06 MB · BSD-3-Clause
Real-ESRGAN x4plus (RRDBNet, 23 blocks)
realesrgan-x4plus · 70.44 MB · BSD-3-Clause
Real-ESRGAN x4plus anime (RRDBNet, 6 blocks)
realesrgan-x4plus-anime · 18.84 MB · BSD-3-Clause
Real-ESRGAN realesr-general-x4v3 + wdn (SRVGGNetCompact, DNI 0.5)
realesr-general-denoise-x4v3 · 5.05 MB · BSD-3-Clause
Denoise variant produced locally by Deep Network Interpolation of the two official checkpoints; see spike/export_upscalers_v2.py.
YuNet face detector (libfacedetection)
yunet-face-det · 0.23 MB · MIT
YuNet by Shiqi Yu and Wei Wu, MIT licensed, via OpenCV Model Zoo.
https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet
MODNet photographic portrait matting
modnet-portrait · 25.89 MB · Apache-2.0
U^2-Net
u2net · 176.00 MB · Apache-2.0
PP-OCRv5 mobile rec (Chinese, English, Japanese)
ocr-rec-ch · 16.53 MB · Apache-2.0
PP-OCRv5 mobile rec (Latin scripts)
ocr-rec-latin · 8.04 MB · Apache-2.0
PP-OCRv5 mobile rec (English)
ocr-rec-en · 7.85 MB · Apache-2.0
PP-OCRv5 mobile rec (Cyrillic)
ocr-rec-cyrillic · 8.05 MB · Apache-2.0
PP-OCRv5 mobile rec (East Slavic)
ocr-rec-eslav · 7.89 MB · Apache-2.0
PP-OCRv5 mobile rec (Arabic scripts)
ocr-rec-arabic · 8.00 MB · Apache-2.0
PP-OCRv5 mobile rec (Devanagari)
ocr-rec-devanagari · 7.91 MB · Apache-2.0
PP-OCRv5 mobile rec (Korean)
ocr-rec-korean · 13.42 MB · Apache-2.0
PP-OCRv5 mobile rec (Thai)
ocr-rec-th · 7.89 MB · Apache-2.0
PP-OCRv5 mobile rec (Greek)
ocr-rec-el · 7.81 MB · Apache-2.0
PP-OCRv5 mobile rec (Tamil)
ocr-rec-ta · 7.89 MB · Apache-2.0
PP-OCRv5 mobile rec (Telugu)
ocr-rec-te · 7.90 MB · Apache-2.0
Whisper tiny encoder (int8) — OpenAI, ONNX by Xenova
whisper-tiny-encoder-int8 · 10.12 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper tiny decoder (int8) — OpenAI, ONNX by Xenova
whisper-tiny-decoder-int8 · 30.73 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper tiny encoder (fp16) — OpenAI, ONNX by Xenova
whisper-tiny-encoder-fp16 · 16.52 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper tiny decoder (fp16) — OpenAI, ONNX by Xenova
whisper-tiny-decoder-fp16 · 59.60 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper base encoder (int8) — OpenAI, ONNX by Xenova
whisper-base-encoder-int8 · 23.20 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper base decoder (int8) — OpenAI, ONNX by Xenova
whisper-base-decoder-int8 · 53.71 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper base encoder (fp16) — OpenAI, ONNX by Xenova
whisper-base-encoder-fp16 · 41.33 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
Whisper base decoder (fp16) — OpenAI, ONNX by Xenova
whisper-base-decoder-fp16 · 104.74 MB · Apache-2.0
Whisper by OpenAI (MIT upstream); ONNX conversion Xenova/transformers.js (Apache-2.0 tag).
MI-GAN inpainting (Picsart AI Research)
migan · 29.55 MB · MIT
MI-GAN by Sargsyan et al. (ICCV 2023), MIT licensed; ONNX export by the first author.
HTDemucs v4 (Meta AI Research) — music source separation
htdemucs · 90.56 MB · MIT
Demucs by Alexandre Défossez, Simon Rouard et al. (Meta AI), MIT licensed; ONNX export of the network body by this site.
AudioSeal 16-bit generator (Meta AI Research)
audioseal-gen · 29.55 MB · MIT
AudioSeal by San Roman et al. (Meta AI, ICML 2024), MIT licensed code and weights.
AudioSeal 16-bit detector (Meta AI Research)
audioseal-det · 17.39 MB · MIT
AudioSeal by San Roman et al. (Meta AI, ICML 2024), MIT licensed code and weights.
Libraries
The open-source libraries that do the non-ML work, shipped to your browser as part of this site. Notices travel with the redistribution, as their licences ask.
mediabunny · MPL-2.0 · video/audio demux, decode, encode, mux
https://github.com/Vanilagy/mediabunny
@mediabunny/mp3-encoder · MPL-2.0 · MP3 encoding (WASM)
https://github.com/Vanilagy/mediabunny
Bundles a WebAssembly build of the LAME MP3 encoder (LGPL-2.1). Source: https://lame.sourceforge.io/
pdf.js · Apache-2.0 · PDF rendering and text extraction
https://github.com/mozilla/pdf.js
onnxruntime-web · MIT · ONNX model inference (WebGPU/WASM)
https://github.com/microsoft/onnxruntime
fflate · MIT · ZIP reading and writing
https://github.com/101arrowz/fflate
gifenc · MIT · GIF quantisation and encoding
https://github.com/mattdesl/gifenc
@cantoo/pdf-lib · MIT · PDF merge/split/reorder and encryption
https://github.com/cantoo-scribe/pdf-lib
libheif · LGPL-3.0-or-later · HEIC/HEIF decoding (WASM, via libheif-js)
https://github.com/strukturag/libheif
Loaded as a separate WebAssembly module; complete source at the link.
UTIF.js · MIT · TIFF decoding
https://github.com/photopea/UTIF.js
libarchive.js · MIT (libarchive: BSD-2-Clause) · RAR/7z/TAR extraction (WASM)
https://github.com/nika-begiashvili/libarchivejs
Signalsmith Stretch · MIT · pitch shifting and time stretching (WASM)
https://github.com/Signalsmith-Audio/signalsmith-stretch
Copyright (c) 2022 Geraint Luff / Signalsmith Audio Ltd. Vendored from npm [email protected]; see src/vendor/signalsmith-stretch.