Introduction

Pruna AI just launched P-Video-2-Pro-Avatar, a talking-video model built on MiniMax H3. Give it one portrait plus either a text script or your own audio file, and it returns a lip-synced 24 fps video with natural expressions, anywhere from 1 second to 3 minutes long. It's a closed, API-only model (no open weights), and you can run it today through Pruna's API and Playground or through partners like Replicate, Together, WaveSpeed, Cloudflare, and Scenario.

If you read our earlier piece on P-Video-2-Pro, think of this as its chatty sibling: same H3 backbone, now tuned to make one still face talk. ✨

What shipped

  • Model: P-Video-2-Pro-Avatar (API model id p-video-2-pro-avatar)
  • Vendor: Pruna AI, built on MiniMax H3
  • Modality: image plus speech to talking video
  • Access: closed model via API, with a rate limit of 50 requests per minute
  • Inputs: one portrait (JPG, PNG, or WEBP up to 20 MiB) plus either uploaded audio (up to 64 MiB and 3 minutes) or a voice_script
  • Output: 24 fps MP4, 1 second to 3 minutes, at 480p or 768p natively, with 1080p, 2K, and 4K made by upscaling the 768p render
  • Voices: 30 built-in text-to-speech voices across 10 languages, including English (US and UK), Spanish, French, German, Japanese, Korean, and Hindi
  • Controls: voice delivery prompt, scene prompt, seed, and prompt upsampling

The audio sets the length. If you upload a track, the video matches it exactly, and if you send both audio and a script, the uploaded audio wins.

Pricing

Pruna is running 50% off for the first two weeks, until November 5 at 18:00 CET. During that window the docs list:

  • 480p: $0.02 per second of output
  • 768p: $0.04 per second of output
  • 1080p: $0.06 per second
  • 2K: $0.08 per second
  • 4K: $0.12 per second, capped at 125 seconds of audio

That means a full one-minute talking clip costs $1.20 at 480p or $2.40 at 768p. Voice, language, and prompt upsampling don't change the price.

Speed

Pruna quotes roughly 0.97 seconds of compute per second of output at 480p and about 2.2 seconds at 768p. So a 480p clip renders in about real time, which is handy for batches of explainer or UGC-style videos.

Where to run it today

Pruna API

Upload the portrait, then post a prediction with a script. The full reference lives in the P-Video-2-Pro-Avatar docs.

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: $PRUNA_API_KEY" \
  -F "content=@portrait.jpg"

curl -X POST "https://api.pruna.ai/v1/predictions" \
  -H "Content-Type: application/json" \
  -H "apikey: $PRUNA_API_KEY" \
  -H "Model: p-video-2-pro-avatar" \
  -d '{
    "input": {
      "image": "https://api.pruna.ai/v1/files/YOUR_FILE_ID",
      "voice_script": "Hello, welcome to our product demo!",
      "voice": "Zephyr (Female)",
      "voice_language": "English (US)",
      "resolution": "768p"
    }
  }'

Partners and ComfyUI

Pruna lists Replicate, eachlabs, Cloudflare, Scenario, WaveSpeed, WiroAI, Sociaro, Tellers, and Together as inference partners. ComfyUI also quoted the launch with "Pruna P-Video-2-Pro Avatar will be available in ComfyUI!", so a Partner Node is on the way, though it isn't in a tagged release yet.

Tips for better takes

  • Prompt upsampling is on by default. A vision model reads your portrait and adds the scene's own motion, like rain, lights, or body movement, after your prompt. Set prompt_upsampling to false when you want your prompt used exactly as written.
  • The portrait sets the aspect ratio. Crop it the way you want the final frame to look before you upload.
  • Use voice_prompt for delivery. It steers tone, pacing, accent, and emotion without being spoken aloud.

Here's a scene prompt to start from:

The woman speaks warmly to the camera, smiling between sentences, soft window light flickering across her face, a gentle breeze moving her hair.

Original Source

Conclusion

P-Video-2-Pro-Avatar makes one portrait plus one script a complete talking video, at prices low enough to try a dozen takes during the launch discount. Grab your best headshot, write a short hello, and see how your face sounds in H3. Want to play with MiniMax H3 for your own clips too? Try this on Gen → https://artrealmai.com/gen?utm_source=magazine&utm_campaign=gen&utm_content=pruna-p-video-2-pro-avatar-minimax-h3-talking-portrait

—Aurelia ♡