Pruna P-Video-2-Pro-Avatar: one portrait plus audio or a script becomes a talking video, built on MiniMax H3

Pruna AI announced P-Video-2-Pro-Avatar on Oct 22. Its launch post passed 600 likes and roughly 100k views within hours, and @ComfyUI quoted it to say the model 'will be available in ComfyUI'. Pruna describes it as based on MiniMax H3. The input is a single portrait plus either an uploaded audio track or a text script. With a script, it generates the speech from 30 voices across 10 languages. Output runs 1 s to 3 min at 24 fps.

Pricing is per second of output. The launch post lists $0.02/s at 480p and $0.04/s at 768p, both 50% off for two weeks (until Nov 5, 18:00 CET). Pruna quotes about 0.97 s of compute per output second at 480p and about 2.2 s at 768p. The docs also list 1080p, 2K and 4K tiers. Those are upscales of the 768p render, and 4K is capped at 125 s of audio. The controls are voice delivery (a voice prompt), scene motion, seed, and prompt upsampling. Upsampling is on by default and uses a vision model to add scene motion to your prompt.

You can reach it through Pruna's own API and playground, or through the partners it lists: Replicate, eachlabs, Cloudflare, Scenario, WaveSpeed, WiroAI, Sociaro, Tellers and Together. For a faceless channel or explainer, that means a presenter shot can be one API call that takes your finished voiceover, instead of a separate lip-sync pass. At 768p the launch price works out to about $2.40 per minute of output. That is our arithmetic from Pruna's price, not a quote.

Use your own face, a fictional character, or someone who has given written permission. The docs list a safety filter that is off by default, so turn it on if you put this behind a user-facing app. We haven't run it yet. Check how it handles long scripts, hands and profile angles before you rely on it, and compare it with the MiniMax H3 ComfyUI workflows in our directory.