Tavus' Griffin video persona: near-call latency, best used as a tireless tutor
sirajraval · x · 2026-10-22
- After a week with Tavus' new video persona Griffin, the author calls it the first AI he's talked to face to face that actually holds a conversation — latency close enough to a real call that he stopped noticing it.
- The web interface shows a two-way video call with preset interaction prompts (thumbs up, copy me, mid-conversation cut-ins) supporting real-time interruption.
- His best use so far: a language tutor that corrects you mid-sentence and never gets tired. More to come.
More from Multimodal
- Scoble: AI Interactive Ideograms Feel Like AI Video a Few Years Ago — Scobleizer · 2026-10-22
- Kling 4.0 Hands-on: 30-Second Videos With Multi-Image, Audio References — aziz4ai · 2026-10-22
- Voxelo 3.0 launches on PlayCanvas Gaussian splats with SuperSplat publishing, targets AI shopping traffic — willeastcott · 2026-10-22
- 'The Rescue' short film made with Seedance 2.5 on Runway, prompt included — umesh_ai · 2026-10-22
- Fake Transylvania tourist photo with Magnific: creator shares the prompt — techhalla · 2026-10-22
- DuET turns piano recordings into engraved sheet music with chords and key signatures — linoy_tsaban · 2026-10-22