Practical overview
When it fits
Use Dragdiffusion when you want to explore or implement an interactive diffusion-editing workflow that directs changes with masks and point positions, and can run the documented local Python interface.
Who it is for
- Researchers studying diffusion-based image editing.
- Developers prototyping interactive, point-guided image manipulation.
Documented uses
- Guide an image edit by marking an editable region and moving image content between chosen points.
- Train a LoRA from an input image and prompt as part of the real-image editing workflow.
- Generate an image through the interface and then edit it with mask and point controls.
Requirements & limits
The repository is public and its algorithm code is licensed under Apache 2.0. The project is presented as research code, not a commercial product.
- Local setup uses Conda and the repository’s environment.yaml file.
- The maintainers recommend Linux with an Nvidia GPU and report around 14 GB of GPU memory; they had not tested other configurations.
- The real-image workflow trains a LoRA before editing; the generated-image workflow first creates an image using generation parameters.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the local editing interface
Create the Conda environment and start Dragdiffusion’s Gradio interface.
Before you start
- Python code checked out from the public repository.
- Conda to create and activate the environment described by environment.yaml.
- The maintainers recommend Linux with an Nvidia GPU and report around 14 GB of GPU memory; other configurations have not been tested.
Steps
Create the environment with
conda env create -f environment.yaml.Activate it with
conda activate dragdiff.Start the interface with the command below.
python3 drag_ui.py
Expected outcome
The README’s local Gradio interface is started for the documented image-generation and point-editing workflows.
Guide limitations
- The project recommends Nvidia GPU hardware on Linux and says it requires around 14 GB of GPU memory.
- The maintainers had not tested other configurations.
Edit an input image with mask and point controls
Use the interface workflow to guide an edit to an input real image.
Before you start
- The Dragdiffusion interface started locally.
- An input image and a prompt describing it.
Steps
Load the input image and enter its descriptive prompt.
Use the interface’s Train LoRA control to train a LoRA for the image.
Draw a mask over the editable area.
Set handle and target points, then run the algorithm to display the edited result.
Expected outcome
The interface displays an edited result guided by the selected mask and points.
Guide limitations
- This recipe follows the documented input-image workflow; the project recommends Linux with an Nvidia GPU and around 14 GB GPU memory.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- Yujun-Shi/DragDiffusion · Official homepage title · retained in its source language