p-video-2

Generate premium AI videos from text, images, or audio with native-speech lip-sync, sharp close-ups, and strong identity consistency.

P-Video-2 is Pruna’s quality-focused video generation model and the successor to P-Video.

How it works

The model takes:

• A prompt describing the subject, action, scene, and optionally camera, lighting, style, and audio.
• An optional image for image-to-video generation, or audio for audio-conditioned generation.
• A resolution, frame rate, duration, aspect ratio, and draft setting.

It returns a generated video at up to 1080p and 48 fps, with optional native audio.

Tips

• Start with the subject, action, and scene. Add camera movement, lighting, style, and dialogue when you need more precise and repeatable results.
• Turn draft on for faster, lower-cost iteration, then turn it off for final-quality generations.
• For native speech and lip-sync, write the dialogue directly in the prompt, keep save_audio on, and don’t provide an audio input.
• Use an audio input when generation needs to follow a specific track. The output duration follows the audio, so duration is ignored.
• For image-to-video, use a high-quality, well-lit reference image and keep the requested motion consistent with the starting frame. aspect_ratio is ignored when an image is provided.
• Use last_frame_image when you need control over the end frame.
• Leave duration empty to let the model choose the video length from the prompt, or set it explicitly from 1 to 20 seconds.
• Prefer 24 fps at 1080p for stability, or 48 fps at 720p for smoother motion.

Resources

Docs: https://docs.pruna.ai/en/stable/docs_pruna_endpoints/index.html