Built with open-source models

TinyModel uses public models and browser runtimes. Models are downloaded as static assets and inference takes place on your device.

• FFmpeg.wasm core 0.12.10 — GPL-2.0-or-later, including the enabled codec libraries; used for local MOV to MP4 conversion. This is not the MIT-licensed wrapper. See the corresponding source and build scripts and GPL license.
• BiRefNet Lite 512 — MIT model card; based on BiRefNet.
• PP-OCRv6 Small detection and recognition — Apache-2.0.
• Whisper — MIT; ONNX conversion by ONNX Community.
• ESRGAN Slim and UpscalerJS — MIT.
• PaddleOCR.js — Apache-2.0; the official browser OCR pipeline.
• UTIF.js 3.1.0 — MIT; bundled into the image-conversion Worker to decode TIFF files locally. See the UTIF and pako license notices.
• libheif-js 1.23.2 — LGPL-3.0; its verified WebAssembly decoder is loaded only when an image needs HEIC/HEIF decoding. See the license and libheif source.
• The HEIC sample shown in the image conversion guide comes from the libheif examples and is distributed under MIT; see its license.
• Transformers.js, ONNX Runtime Web, OpenCV, Clipper, Next.js, React, MDX, and Lucide support the application.

Exact model revisions and file checksums are recorded in the project model lockfile. License texts accompany the repository's third-party notices. The decorative woodland illustration was generated specifically for TinyModel.