November 25, 2025
Five legs, zero chill
Meta Segment Anything Model 3
Finds anything—even 5 dog legs—fans hype, skeptics say "old news"
TLDR: Meta released SAM 3 and SAM 3D with open model weights and an easy playground, letting anyone find and track objects by text prompts. Comments oscillate between wow (it spotted a five-legged dog) and meh (old news, can it run on a Raspberry Pi), highlighting real-world creative and shopping impact.
Meta just dropped SAM 3, an AI that can spot, outline, and track whatever you describe—think “the striped red umbrella,” not just “person.” It’s bundled with an open playground, model weights, and a 3D companion that rebuilds scenes from a single photo. But the internet didn’t just nod; it exploded. One user ran the infamous 5-legged dog test and claimed SAM cleanly marked all five—while other “smart” systems still insist dogs only have four. Cue jokes about “bonus legs” and whispers that Meta’s onto a future super-brain.
Meanwhile, the hype squad is swooning over SAM 3D, saying it rigs multiple humans from one pic and nails tricky objects like translucent lamps. Marketplace’s new “View in Room” and creator toys for Instagram’s Edits app feed the “this is actually useful” crowd. But the snark brigade chimed in: “Wasn’t this on the front page last week?”—translation: great, but don’t act brand-new.
Pragmatists crashed the party asking for captioning tools and whether it can run on a Raspberry Pi (aka “can it run on my toaster?”). The vibe: split between “this is a creative revolution” and “calm down, it’s segmentation.” Yet with open weights, a playground, and a wildlife dataset, even skeptics admit—this drop has teeth.
Key Points
- •Meta launched Segment Anything Model 3 (SAM 3), a unified model for detection, segmentation, and tracking using text, exemplar, and visual prompts.
- •Meta released SAM 3 model weights, evaluation datasets, fine-tuning code, a demo on Segment Anything Playground, and a research paper.
- •SAM 3 introduces promptable concept segmentation to find and segment instances defined by open-vocabulary text or exemplar prompts.
- •Meta introduced SAM 3D, an open-source suite for single-image 3D object/scene reconstruction and human pose/shape estimation.
- •SAM 3 and SAM 3D power Facebook Marketplace’s View in Room feature and will enable new creative tools in Meta AI apps and Instagram’s Edits; Meta also launched a wildlife-monitoring video dataset with Conservation X Labs and Osa Conservation.