pruna / p-video-avatar

## Overview P-Video-Avatar turns one portrait image into a speaking video clip. You provide a script for text-to-speech or upload your own audio. The model then animates the face to match the speech and adds natural head motion. It’s useful when you need a consistent on-camera “spokesperson” without filming. ## What you can build - Product explainers with a consistent presenter - Personalized video messages from a single headshot - Multilingual announcements using the supported voice languages - Talking character portraits for social content - Audio-driven avatar replies from customer support recordings ## Inputs - A single portrait image of the person to animate, as JPG, JPEG, PNG, or WebP. Use a clear, front-facing photo for best lip sync. - A short script for the person to say, provided as plain text. Use this only when you don’t upload audio. - A preset synthetic voice selection for the generated speech. This is ignored when you upload audio. - A speech output language selection for the synthetic voice. This is ignored when you upload audio. - An optional audio file to drive the lip sync, as FLAC, MP3, or WAV. When you upload audio, it takes priority over the script. - A required output resolution choice: 720p or 1080p. - An optional video description prompt that nudges the motion and scene behavior, such as “the person is talking” or “calm, minimal movement.” - An optional speaking-style prompt that guides tone, pacing, and emotion for generated speech. - An optional random seed value to help reproduce similar results across runs. - An option to enable or skip automated safety checks on the prompt and image before generation. - An option to enable or skip automatic prompt upsampling before the video is generated. ## Outputs The model returns a generated video file (typically MP4) showing your portrait animated into a talking-head clip. The video matches the selected resolution (720p or 1080p). If you provide a script, the output includes a synthesized speech track. If you upload audio, the output uses your audio for timing and lip sync. ## Limitations - The input image must be a portrait. Side profiles, occluded mouths, masks, hands over the face, or extreme angles reduce lip-sync quality. - Low-quality inputs can break identity consistency. Blurry photos, heavy filters, compression artifacts, or tiny faces often cause jitter or warped features. - The voice and language choices apply only to script-based speech generation. Uploaded audio overrides them. - Prompt upsampling can change the meaning of short prompts. If you need strict control, disable it. - Disabling safety checks increases the risk of generating disallowed content. Use it only when you have strong moderation. ## Safety & compliance Follow applicable consent and right-of-publicity rules before generating a real person’s likeness. Don’t create misleading or deceptive content, including impersonation meant to defraud. Respect copyright for any images and audio you upload. If you disable safety checks, you’re responsible for ensuring the content complies with platform policies and local laws.