Instructions to use SceneWorks/sam3-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use SceneWorks/sam3-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir sam3-mlx SceneWorks/sam3-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
SceneWorks/sam3-mlx
A redistribution mirror of facebook/sam3
(Segment Anything 3) for download-on-first-use by the SceneWorks macOS worker.
The native-MLX (Rust mlx-gen-sam3) SAM3 segmenter loads the stock checkpoint directly โ no
conversion โ and drives text-concept Promptable Concept Segmentation ("person") for the
Replace-Person person-track pipeline (epic 4910). This mirror exists only so the app can fetch the
weights without the gated facebook/sam3 access flow.
Files
model.safetensorsโ the SAM3 checkpoint (~0.9B params, F32), byte-identical tofacebook/sam3.tokenizer.jsonโ the CLIP-H text tokenizer used for the concept prompt.LICENSEโ the Meta SAM License (included as required by the license; it governs use of the Materials).
License
These weights are ยฉ Meta and distributed under the SAM License (see LICENSE). This mirror
redistributes them unmodified under that license. Refer to the original
facebook/sam3 model card for intended use, capabilities,
and limitations.
Model size
0.9B params
Tensor type
F32
ยท
Hardware compatibility
Log In to add your hardware
Quantized
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support