Generate talking head avatar videos with Pruna P-Video-Avatar via inference.sh CLI. Turn a portrait image into a realistic speaking video with built-in TTS. 18x faster and 6x cheaper than competitors. Models: P-Video-Avatar, P-Image (for portrait generation). Capabilities: text-to-avatar, audio-driven avatars, 30 voices, 10 languages, 720p/1080p, built-in TTS, dynamic backgrounds, full-body control. Use for: AI presenters, product demos, explainer videos, virtual influencers, marketing, educatio
The p-video-avatar skill generates talking-head avatar videos from a single portrait image using Pruna P-Video-Avatar through the inference.sh CLI (belt). It turns a portrait into a realistic speaking video with built-in text-to-speech, positioned as the fastest and most cost-effective avatar model available — described as quality on par with Veo 3.0 but roughly 18x faster and 6x cheaper than alternatives like Fabric, OmniHuman, and HeyGen. The core workflow is belt login followed by belt app run pruna/p-video-avatar with a JSON --input containing an image plus either a voice_script or an audio file.
Capabilities include text-to-avatar and audio-driven avatars, 30 voices, 10 languages, 720p/1080p output, dynamic backgrounds, and full-body control. Parameters include image (required portrait), voice_script, audio (which overrides voice_script when both are present), voice (default "Zephyr (Female)"), voice_language (default "English (US)"), resolution, video_prompt to control avatar behavior and background, voice_prompt to control tone, pacing and emotion, seed for reproducibility, disable_safety_filter, and disable_prompt_upsampling. Available voices are listed by gender (female voices such as Zephyr, Kore, Leda, Aoede; male voices such as Puck, Charon, Fenrir, Orus), and supported languages include English (US/UK), Spanish, French, German, Italian, Portuguese (Brazil), Japanese, Korean, and Hindi. A documented full workflow uses pruna/p-image to generate a 9:16 portrait first, then animates it. Pricing is $0.025 per second at 720p and $0.045 per second at 1080p, with a stated free launch weekend from May 1 to May 4, 2026 (CET).
The skill targets marketers, educators, and creators making AI presenters, product demos, explainer videos, virtual influencers, multilingual localization, and corporate or gaming content. It requires the inference.sh CLI (belt) via npx skills add belt-sh/cli and limits its allowed tool to Bash(belt *). Tips note the output aspect ratio matches the input image and recommend keeping videos under 3 minutes for best visual consistency.
A required portrait image (jpg, jpeg, png, or webp) plus either a voice_script (text for the person to say) or an audio file. When both audio and voice_script are provided, audio takes priority.
30 voices (listed as female voices like Zephyr, Kore, Leda, Aoede and male voices like Puck, Charon, Fenrir, Orus) and 10 languages including English (US/UK), Spanish, French, German, Italian, Portuguese (Brazil), Japanese, Korean, and Hindi.
Pricing is $0.025 per second of output video at 720p and $0.045 per second at 1080p (so a 30-second 720p video is $0.75). The doc states P-Video-Avatar is completely free from Thursday May 1, 2026 4:00 PM CET through Sunday May 4, 2026 11:59 PM CET.
Use video_prompt to control avatar behavior, body language, and dynamic backgrounds, and voice_prompt to control tone, pacing, and emotion. Resolution can be set to 720p or 1080p.
The inference.sh CLI called belt, installed via npx skills add belt-sh/cli, then belt login. The allowed tool is Bash(belt *). To create a portrait first, the skill suggests pruna/p-image with aspect ratio 9:16 for vertical avatar videos.
Quick Setup:
.claude/skills/Repository
101-skills/skills