Is persona-2.0-live state of the art?Pending benchmarks.Announced Oct 12, 2026; no independent results yet.
Keyframe Labs' real-time avatar model with word-level control over facial expression. First offered in preview on October 12, 2026, and generally available on every plan as of November 12, 2026. Official pages list eleven directed emotions, 512×512 frames, about 180 ms model latency, and pricing from $0.06 per minute (6 credits per minute). Available through the Keyframe platform, API, and a homepage demo. No public weight release is stated.
The details so far
- Lab
- Keyframe Labs
- Announced
- Oct 12, 2026
- API
- Live · persona-2.0-live
- Open weights
- Closed · No public weight release stated on the official launch post, changelog, or docs
- Type
- Real-time avatar model that turns streaming audio into lip-synced video, with word-level control over facial expression via emotion tags or SDK cues (Keyframe launch post and docs)
- Controls
- Official launch post: eleven emotions (neutral, happy, sad, angry, surprised, concerned, confused, disgusted, excited, scared, and skeptical). Each switch is described as landing on its word, within about 40 ms. Frame size is 512×512, versus 256×256 for persona-1.5-live; model latency is about 180 ms (Keyframe launch post)
- Hosted access
- Generally available on every plan as of November 12, 2026, through the Keyframe platform and API, with a homepage demo. Official preview access began October 12, 2026. API keys are created in the Keyframe platform dashboard (Keyframe GA changelog, launch post, and docs)
- Pricing
- Official launch post: 6 credits per minute, with extra minutes from $0.06 per minute, the same price as persona-1.5-live (Keyframe launch post and changelog)
- License
- Not stated as a weight license. Access is under Keyframe Labs terms
- Related
- Separate from persona-1.5-live, the previous generally available Keyframe avatar model
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: self-reported EmoteBench results from a blinded study of 250 participants and 19,500 judgments across five real-time avatar APIs: 81% correct emotion identification versus 37% for Tavus (2.2 times), 61% rated very or completely natural versus 32% for Anam (1.9 times), and 53% correct identification of both emotions in a mid-sentence switch versus 13% for Tavus (4 times) (Keyframe launch post).
Sources
LABPersona-2.0-Live in preview (Keyframe Labs changelog)keyframelabs.com/changelog/persona-2-live · Oct 12, 2026LABPersona-2.0-Live generally available (Keyframe Labs changelog)keyframelabs.com/changelog/persona-2-live-ga · Nov 12, 2026LABPersona-2.0-Live (Keyframe Labs)keyframelabs.com/blog/persona-2-live · Nov 12, 2026LABParth Radia on X: persona-2.0-livex.com/parthnradia/status/2120939721560059963 · Nov 12, 2026Tracking continues until an independent board lists it.