snap-research/stable-flow
Research implementation of a training-free image editing method using diffusion transformers and flow matching for controlled image modifications.
Not currently ranked — collecting fresh signals.
star history
This repository provides the official implementation of a CVPR 2025 paper on image editing using diffusion transformers. The method identifies vital layers within DiT architectures for selective attention feature injection, enabling consistent edits ranging from non-rigid modifications to object addition. It introduces an improved image inversion method for flow models to enable real-image editing, and includes qualitative and quantitative evaluation with a user study.
Frequently asked
- What is snap-research/stable-flow?
- Research implementation of a training-free image editing method using diffusion transformers and flow matching for controlled image modifications.
- Is stable-flow open source?
- Yes — snap-research/stable-flow is an open-source project tracked on heatdrop.
- What language is stable-flow written in?
- snap-research/stable-flow is primarily written in Python.
- How popular is stable-flow?
- snap-research/stable-flow has 409 stars on GitHub.
- Where can I find stable-flow?
- snap-research/stable-flow is on GitHub at https://github.com/snap-research/stable-flow.