p-video

p-video is a skill for generating videos with Pruna's optimized P-Video and WAN models through the inference.sh `belt` command-line tool. Pruna optimizes video models for speed and lower cost while preserving quality, and this skill packages that catalogue so an agent can turn a text prompt or a still image into a finished video clip with one shell command, avoiding local GPU setup or SDK integration.

It documents three models: P-Video for text-to-video and image-to-video with optional synchronized audio input, WAN-T2V for economical text-to-video at 480p/720p, and WAN-I2V for animating still images. Each runs as `belt app run pruna/<model> --input '{...}'` with JSON parameters for prompt, duration, resolution (720p/1080p depending on model), an input image or audio URL, and a draft mode for faster, cheaper previews. The skill includes a pricing table (for example WAN-T2V at roughly $0.05 for 480p and $0.10 for 720p per video) and links to related inference.sh skills for image generation, image-to-video, and text-to-speech narration.

Typical users are content creators, marketers, and developers building media pipelines who want fast, low-cost AI video — animated stills, short social clips, concept tests, and cinematic landscape shots. Like its sibling image skill, it is a documented front-end to the hosted inference.sh service rather than a standalone video engine, so its capabilities are bounded by what that platform offers.