Skip to content

TCAN

Portrait animation with temporal consistency

Visit official site

What TCAN is

The framework is described as a diffusion-model-based portrait animation approach that keeps temporal consistency and generalises to unseen domains. Its named modules keep the source appearance, follow the driving pose and preserve the background of the source image.

What you can do with it

  • Keep the source appearance while retaining pose information through the APPA layer
  • Use a temporal control network to prevent collapse from abrupt pose changes
  • Apply a pose-driven temperature map to reduce flicker in static regions
  • Maintain temporal consistency, background consistency and generalisation to unseen domains

Application scenarios

  • Generate character action shots in film production
  • Build animated character motion in game development
  • Produce virtual character animation for interaction in VR

Getting started

  • Prepare a source image and a driving video with the needed appearance and motion
  • Feed both inputs into the model to generate portrait animation
  • Adjust parameters such as APPA layer weight and control network strength
  • Use the pose-driven temperature map to reduce flicker and discontinuity

Pros & cons

✓ What we like

  • The APPA layer keeps appearance while retaining pose information
  • A temporal control network and a temperature map address pose jumps and flicker

! What to watch out for

  • Generation takes both a source image and a driving video as inputs
  • Parameters such as APPA layer weight and control network strength are adjusted during generation

FAQ

Which modules does the framework name?

The material lists the APPA layer, the temporal control network and the pose-driven temperature map.

What inputs are needed?

A source image and a driving video that carry the needed appearance and motion.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review