Skip to content

Stable Video Portraits

Generate dynamic face videos from a 2D/3D hybrid method

Visit official site

What Stable Video Portraits is

Stable Video Portraits is described as a hybrid 2D/3D generation method that lifts a general 2D diffusion model into a video model through person-specific fine-tuning. Time-series 3D face shapes act as conditioning and a temporal denoising process gives smooth face imagery.

What you can do with it

  • Generate realistic dynamic face videos
  • Extract a 3D face reconstruction for each input frame
  • Condition generation on time-series 3D face shapes
  • Morph a personal avatar into a text-defined celebrity

Application scenarios

  • Generate realistic virtual characters for film production
  • Create highly realistic NPC face animations for games
  • Produce distinctive digital artwork

Getting started

  • Visit the official site of the project
  • Read the research paper and method overview
  • Download and install the required software and libraries
  • Prepare input video that suits 3D face reconstruction

Pros & cons

✓ What we like

  • Head pose is kept during face morphing
  • Input video is handled with 3D face reconstruction and Mediapipe

! What to watch out for

  • The method builds on a pretrained text-to-image diffusion model
  • Input video must meet 3D face reconstruction requirements

FAQ

What does it generate?

The material describes realistic dynamic face videos with temporal smoothness.

How is the input video processed?

The material names 3D face reconstruction, a face parsing map model and Mediapipe.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review