Skip to content

InstantDrag

Edit images with drag instructions and no optimisation

Visit official site

What InstantDrag is

The method is described as an optimisation-free pipeline for drag-based image editing that takes only an image and drag instructions. It splits the task into motion generation and motion-conditioned image generation.

What you can do with it

  • Drag a region to move an object in the image
  • Let FlowGen estimate dense optical flow
  • Apply FlowDiffusion to edit the original image

Model notes

  • FlowGen uses a Pix2Pix framework for sparse-to-dense flow
  • FlowDiffusion is conditioned on the image and downsampled flow
  • Training uses the CelebV-Text face video dataset

Pros & cons

✓ What we like

  • Editing is described as fast and realistic
  • Motion dynamics are learned from real-world video data

! What to watch out for

  • Input is limited to an image and drag instructions
  • Training used the CelebV-Text face video dataset

FAQ

What are the two networks?

FlowGen generates optical flow and FlowDiffusion performs flow-conditioned image generation.

Does it need a text prompt?

The material states no mask or text prompt is required.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review