RepFusion: Leveraging Multimodal Priors for Denoising in Representation Space

Benchmark Model Rank Results
image-generation-on-wiseRepFusion-SFT w/ Diffusion Decoder–Overall: 0.64
text-to-image-generation-on-dpg-benchRepFusion-SFT w/ Diffusion Decoder–DPG-Bench Score: 85.11
text-to-image-generation-on-genevalRepFusion-SFT w/ Diffusion Decoder–Overall: 0.87