P-Video 2 Pro Avatar (Lip-Synced Video)

Overview

P-Video 2 Pro Avatar by Pruna turns a single portrait into a speaking video with synced lip motion. You provide one face image, then drive speech with either uploaded audio or a typed script. The model returns an MP4 with the voice baked into the soundtrack at 24 fps, which solves “presenter on camera” videos without filming.

What you can build

• Product announcements with a consistent on-brand spokesperson
• Customer support clips that speak a fixed script in multiple languages
• Training and onboarding videos with the same host across modules
• Localized marketing variants from one approved portrait
• Quick founder-style updates for social posts and landing pages
• Narrated slides or screen recordings when you only need a talking head insert

Inputs

• A single portrait image used as the first frame (JPG, PNG, or WebP). The image can be up to 20 MB. The image also sets the output aspect ratio.
• Either a typed speaking script (optional). If you choose this path, the model generates speech from text.
• Or an uploaded speech audio file (optional). Audio must be between 1 second and 3 minutes, up to 64 MB. If you provide audio, the model uses it instead of the typed script.
• A voice choice for script-driven speech (optional). This is ignored when you upload speech audio. The voice list includes about 30 named voices.
• A language choice for script-driven speech (optional). This is ignored when you upload speech audio. Supported options include English (US/UK), Spanish, French, German, Italian, Portuguese (Brazil), Japanese, Korean, and Hindi.
• Optional delivery direction for the voice (tone, pacing, accent, emotion). The model uses this as guidance and does not speak it aloud.
• Optional visual direction for what the shot should look like while the person talks. If you leave it blank, the default behavior is a simple “person talking” scene.
• Optional prompt upsampling toggle to add motion cues inferred from the portrait, beyond your visual direction.
• Output resolution selection (480p or 768p). The output frame rate stays fixed at 24 fps.
• Optional seed value to repeat results more reliably across runs.

Outputs

The model returns a single MP4 video file.

• Video includes an audio track with the spoken voice.
• Frame rate is fixed at 24 fps.
• Resolution is either 480p or 768p, based on your selection.
• Duration matches the speech length. With uploaded audio, the video length follows the audio.
• The portrait you provide acts as the opening frame and sets the canvas shape.

Limitations

• You must provide a usable portrait. Strong angles, occlusion, or tiny faces can reduce identity match and lip sync.
• Noisy, clipped, or reverberant audio can cause weaker mouth timing and less natural facial motion.
• The model targets a single speaking subject. Group photos and multi-face portraits can confuse face selection.
• The output aspect ratio follows the input portrait. You can’t force a different canvas without changing the portrait.
• Very long scripts can sound less natural with text-to-speech. If cadence matters, upload real speech audio.

Safety & compliance

• Only use portraits and voices you have rights to use. Get consent for any identifiable person.
• Don’t create deceptive “real person said this” content. Add clear disclosure when viewers might be misled.
• Avoid impersonation, defamation, or misleading political and financial claims.
• Respect copyright for any input images and any scripts you did not author.