Hoksomax/sam3-text-segmentation

SAM3 — Segment Anything with Text

Demo of facebook/sam3, Meta's unified foundation model for promptable segmentation in images and videos.

This Space showcases Promptable Concept Segmentation (PCS): type a short noun phrase ("cat", "red car", "person wearing a hat") and SAM3 detects and masks every instance of that concept in the image.

Built with 🤗 transformers (Sam3Model / Sam3Processor) and Gradio. Also usable as an MCP server — the segment endpoint is exposed as a tool.

Note: facebook/sam3 is a gated model. The Space uses an HF_TOKEN secret from an account that has accepted the license on the model page.