p-video-2-pro-avatar

Turn a portrait and speech into a talking video with Pruna’s p-video-2-pro-avatar.

Give it a photo of a person and either an audio file or a script. It returns an MP4 at 24 fps where the person speaks your words with synced lips and expressions. Videos run from 1 second to 3 minutes, and the speech sets the length.

Use it for product demos, explainers, course videos, localized announcements, or anything else where you need someone on camera saying something specific.

How it works

The model takes:

• An image of the person who speaks. This becomes the first frame.
• Either an audio file or a voice_script.
• An optional prompt describing what the video should show.

If you upload audio, the video follows its timing. If you write a script, the model speaks it with text-to-speech in the voice and language you pick. If you set both, the uploaded audio wins.

The portrait sets the aspect ratio. If the image’s shape is outside what the model supports, it’s cropped to fit.

Inputs

• image (required): portrait of the person who speaks. jpg, jpeg, png, or webp, up to 20 MiB.
• audio: speech audio, up to 64 MiB and 3 minutes. Audio under 1 second is rejected.
• voice_script: the exact words to speak when you don’t upload audio.
• voice: text-to-speech voice for voice_script. 30 voices, each labeled female or male (default Zephyr (Female)).
• voice_prompt: how the voice should sound, such as tone, pacing, accent, or emotion. It isn’t spoken (default Say the following.).
• voice_language: language of voice_script. English (US), English (UK), Spanish, French, German, Italian, Portuguese (Brazil), Japanese, Korean, or Hindi (default English (US)).
• prompt: what the video should show (default The person is talking.).
• prompt_upsampling: let a vision model look at the portrait and add the scene’s own motion, like rain, lights, background movement, or body movement, to your prompt. Your words stay unchanged and come first, and anything you ask to keep still stays still. On by default. Turn it off to use prompt exactly as written.
• resolution: 480p, 768p, 1080p, 2k, or 4k (default 768p). 1080p, 2k, and 4k generate at 768p and then upscale. 4k is limited to 125 seconds of audio.
• seed: set for reproducible results.
• enable_safety_filter: block content in the optional moderation categories. The always-on rules apply either way (default off).

Tips

• Use a clear, well-lit, front-facing portrait where the face is easy to see.
• Upload your own audio when timing matters, like a voiceover you’ve already recorded or a voice clone.
• Use voice_prompt to direct delivery, for example Speak warmly and slowly, like you're explaining something to a friend.
• Use 480p for drafts, then switch to a higher resolution for the final version.
• If the background moves in ways you don’t want, say so in prompt (for example The background stays still.) or turn off prompt_upsampling.

Pricing

You pay per second of output video, based on resolution. Voice, language, and prompt_upsampling don’t change the price.

Resources

Docs: https://docs.api.pruna.ai/guides/models/p-video-2-pro-avatar