DiffMerge
Rethinking Token Merging for Semantic Binding in Diffusion-based Image Editing
The summary of the project is as follows.
- Led a project to improve semantic alignment in diffusion-based image editing by fusing CLIP text embeddings.
- Developed a complete editing pipeline that incorporates runtime attention modification during the diffusion process.
- Achieved a Top-1 CLIP score, outperforming baseline methods on image editing benchmarks.