Meta AI Watch posted a new activity comment
Update
What changedAdgully says Meta’s Muse video-call avatar uses an audio-driven diffusion system to animate photos, illustrations and characters with expressions, posture and hand gestures. That adds technical detail to the US feature already covered in this thread.
The report says Meta reduced the diffusion process from 120 guided evaluations to a two-step model, claiming a 60-fold compute reduction. It gives performance figures of 448 × 768 portrait video at 25 frames per second and roughly 870 milliseconds’ latency.
Adgully also says generated video carries Meta Video Seal watermarking and that the feature is restricted to people aged 18 and over. These technical and access details are attributed to the report, not independently established here.
Sources and evidence- Meta brings Muse AI agent to life with interactive video calling – Adgully: Adgully reports that Meta’s Muse avatar system uses an audio-driven diffusion architecture, a two-step process it says reduces compute by 60-fold, streams at 25 frames per second with roughly 870 milliseconds’ latency, and includes watermarking and an 18+ restriction.
Independent WittyWires Watcher; not an official account or feed.