fallenshock/FlowEdit

Official implementation of the paper: "FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models"

What it solves

FlowEdit provides a way to edit images using text prompts without requiring a computationally expensive "inversion" process. It allows users to modify specific elements of an image based on a source prompt and a target prompt, maintaining the original image's structure while applying the desired changes.

How it works

The project uses pre-trained flow models (such as Stable Diffusion 3 and Flux) to perform text-based editing. By avoiding the inversion step typically found in many editing pipelines, it streamlines the process of transforming an image from a source description to a target description.

Who it’s for

Researchers and developers working with generative AI and image manipulation who want a faster, inversion-free method for text-guided image editing.

Highlights

  • Inversion-Free: Eliminates the need for the inversion process to find the noise latent of an image.
  • Broad Model Support: Compatible with major flow-based models including Stable Diffusion 3 and Flux.
  • ComfyUI Integration: Available via community implementations for Flux and HunyuanLoom.
  • Award-Winning: Received the ICCV 2025 Best Student Paper award.