SnapTop model catalog

The dedicated public model repository for SnapTop, separate from application source and all other account repositories. All SnapTop model downloads originate here. Files are public; no account, login or access token is needed. Models run locally: screenshots, recordings, prompts and voice samples remain on the user's computer.

SnapTop downloads models on demand using the immutable commit and SHA-256 checksums recorded in the application's sam/model-source.json. MODEL_CATALOG.json records file sizes, checksums and exact upstream revisions. SHA256SUMS covers all distributed model payloads. Each model folder includes its original license and provenance. The licenses are model-specific; this repository does not relicense model weights.

SAM3

The root model.safetensors and processor/tokenizer files are unchanged Meta SAM3 weights from facebook/sam3@3c879f39826c281e95690f02c7821c4de09afae7. The root sam3_vit_h.zip is the unchanged ONNX conversion from vietanhdev/segment-anything-3-onnx-models. Both retain the Meta SAM License. The root Transformers files remain a loadable SAM3 snapshot.

Additional models

| silero-vad | MIT | https://pypi.org/project/faster-whisper/1.2.1/ | | perth-watermark | MIT | https://pypi.org/project/resemble-perth/1.0.1/ |

Silero VAD and Perth watermarking are unchanged assets bundled with faster-whisper and resemble-perth. SnapTop downloads and verifies these official public files on demand instead of loading the package-bundled copies. Narration setup prepares Perth with Chatterbox; caption speech detection fetches Silero on first use. No provider account or token is required.

MADLAD is used only to generate application translations during development. Qwen ASR/alignment and Unlimited OCR are optional engines. Software dependencies and runtimes are installed from their official package sources; they are not model payloads in this repository.

Downloads last month
53
Safetensors
Model size
0.9B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support