Skip to content

ConsisID

Identity-preserving text-to-video generation via frequency decomposition

Visit official site

What ConsisID is

ConsisID is described as a text-to-video model that keeps identity through frequency decomposition and frequency-domain control signals.

What you can do with it

  • Generate video consistent with a text description
  • Keep a person's identity stable across frames
  • Encode a reference image and facial landmarks

Getting started

  • Visit the project site or GitHub page
  • Install the required dependencies and model
  • Prepare a text description and a reference image
  • Set parameters following the instructions

Pros & cons

✓ What we like

  • A hierarchical training strategy preserves identity information
  • The model outputs high-quality, identity-preserving video

! What to watch out for

  • Generation needs a text description and a reference image
  • Setup involves downloading dependencies and the model

FAQ

Does it need fine-tuning?

The material states the flow needs no case-by-case fine-tuning.

How is identity kept?

The source describes frequency-aware identity control with global and local face extractors.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review