SAM 3.1 as visual Ctrl+F: text prompt finds 50 bones in an X-ray in 286ms, all local
MaziyarPanahi · x · 2026-10-13
Developer Maziyar Panahi demoed a "Ctrl+F for X-rays" using Meta's SAM 3.1: typing "bone" segments 50 skeletal regions on a hand X-ray in 286 ms, and words like "finger", "hand", "scissors" and "hammer" also yield precise masks on medical images and surgical trays.
How it works: SAM 3.1 takes a word as a prompt and returns a mask for every match, costing roughly 0.17–0.29 s of GPU time per word. The entire pipeline runs locally on a single NVIDIA A100, with the checkpoint (facebook/sam3.1) hosted on Hugging Face. The demo video is the app's own screen redrawn from logged live sessions; the author notes the footage is stock and it's not a medical device.
Related event: SAM 3.1 Enables Ctrl+F for X-Rays, Locating 50 Bones in 286ms(2 posts)→
More from Multimodal
- OBVPM timeline workflow v0.1.4-007 fixes joins, audio sync and multi-shot upscaling for MiniMax H3 — obvpm · 2026-10-13
- ElevenLabs' Recast swaps video characters while keeping lighting, motion, and audio intact — aziz4ai · 2026-10-13
- ElevenLabs' new Recast swaps a character in a shot while keeping every other detail intact — aziz4ai · 2026-10-13
- Paste any Shadertoy shader into ComfyUI: kubakub nodes 0.4 adds shader-to-diffusion workflows — Puzzled_Parking2556 · 2026-10-13
- AI video imagines women from famous paintings living in modern times — cocoweixu · 2026-10-13
- Qwen image LoRA training: are image pairs actually necessary? — Inevitable_Pen9043 · 2026-10-13