P-Video-2-Pro Avatar, released October 22, 2026, turns a portrait plus audio or a script into a talking video from one second to three minutes long. It also adds prompt upsampling, which expands a short instruction into a more detailed generation brief.
Use a portrait to define the performer and an audio recording to supply their delivery. A script is useful when you still need spoken audio; an existing track is the better starting point when the timing and voice are already decided.
Speed and Tellers Token cost
The announced generation speed is approximately 1 to 2.2 seconds of processing per second of output, depending on resolution. That puts a 30-second generation at roughly 30 to 66 seconds of model processing, before queueing, uploads and editing.
| Resolution | Tokens per output second | 30-second avatar | 3-minute avatar |
|---|---|---|---|
| 480p | 0.6 | 18 | 108 |
| 768p | 1.2 | 36 | 216 |
These Tellers Token estimates cover avatar generation. Agent work, separate audio generation, other visuals and additional generations can add to the project total. Request the estimate for your chosen settings before running the full clip.
From a talking portrait to a finished video
In Tellers, the generated avatar becomes footage you can edit. Ask for a tighter opening, change the crop, add captions, or place the performer over other visuals. You can make those changes by prompting or by editing the timeline manually.
The model also appears in our Suno music-video workflow, where a performer is combined with animated lyrics and beat-timed effects. For a more ambitious generated performance, explore Seedance 2.5.
Open Tellers, upload a portrait and your audio, and ask the agent to create an avatar clip with P-Video-2-Pro Avatar.