Skip to content

FLATTEN

Optical flow guidance for consistent video editing

Visit official site

What FLATTEN is

The method is described as an optical flow guided attention plug-in for text-to-video editing. It adds optical flow inside the U-Net of a diffusion model to address consistency.

What you can do with it

  • Add the plug-in to an editing method
  • Edit video without extra training
  • Compare consistency across frames

Key features

  • Optical flow enters the attention modules
  • Matching patches attend across frames
  • Training-free integration is described

Pros & cons

✓ What we like

  • It can join any diffusion-based editing method
  • Visual consistency is the stated focus

! What to watch out for

  • It works as a plug-in for diffusion-based editing methods
  • The paper reports results on existing editing benchmarks

FAQ

Does it need training?

The material states it is training-free and integrates into diffusion-based methods.

What does it improve?

Visual consistency of edited videos across frames.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review