Research map / Image generation / Text-to-image diffusion
Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers
Zhengyao Lv, Tianlin Pan, Chenyang Si, Zhaoxi Chen, Wangmeng Zuo, Ziwei Liu, Kwan-Yee K. Wong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add TACA with LoRA (r=64) | FLUX.1-Dev · T2I-CompBench Spatial | alignment score: 0.2066 -> 0.2405 | Improved |
| add TACA with LoRA (r=64) | FLUX.1-Dev · T2I-CompBench Shape | alignment score: 0.5064 -> 0.5362 | Improved |
| add TACA with LoRA (r=64) | SD3.5-Medium · T2I-CompBench Spatial | alignment score: 0.2087 -> 0.2678 | Improved |
| add TACA with LoRA (r=64) | SD3.5-Medium · T2I-CompBench Shape | alignment score: 0.5770 -> 0.5938 | Improved |
| add TACA with LoRA | FLUX.1-Dev · T2I-CompBench Non-Spatial | alignment score: 0.3035 -> 0.3041 | No clear effect |
| add TACA to FLUX/SD3.5 | FLUX.1-Dev and SD3.5 · MUSIQ / MANIQA image quality | MUSIQ: 0.7186 -> 0.7212 (FLUX), 0.7182 -> 0.7210 (SD3.5) | No clear effect |
| add TACA over LoRA-only fine-tuning (gamma0=1.2) (ablation) | FLUX.1-Dev · T2I-CompBench Color (100 prompts) | alignment score: 0.805 -> 0.839 | Improved |
| add TACA over LoRA-only fine-tuning (gamma0=1.2) (ablation) | FLUX.1-Dev · T2I-CompBench Spatial (100 prompts) | alignment score: 0.187 -> 0.207 | Improved |
| increase gamma0 from 1.2 to 1.3 (ablation) | TACA on FLUX.1-Dev · T2I-CompBench Color/Texture | alignment score: Color 0.839 -> 0.787; Texture 0.790 -> 0.766; Spatial 0.207 -> 0.225 | Mixed |
| vary t_thresh from 970 to 930 and gamma0 from 1.15 to 1.25 (ablation) | TACA · FLUX / SD3.5 attr and spatial | attr / spatial: FLUX attr 0.740 to 0.763 | No clear effect |
| add TACA at CFG scale 10 (ablation) | FLUX.1 Dev · T2I-CompBench Color (100 prompts) | alignment score: 0.667 -> 0.751 | Improved |
| add TACA at CFG scale 7 (ablation) | SD3.5-Medium · T2I-CompBench Color (100 prompts) | alignment score: 0.812 -> 0.843 | Improved |
Closest papers by meaning
- On the Value of Cross-Modal Misalignment in Multimodal Representation Learning (NeurIPS 2025 · similarity 0.88)
- Rare Text Semantics Were Always There in Your Diffusion Transformer (NeurIPS 2025 · similarity 0.88)
- MODA: MOdular Duplex Attention for Multimodal Perception, Cognition, and Emotion Understanding (ICML 2025 · similarity 0.88)
- Towards Transformer-Based Aligned Generation with Self-Coherence Guidance (CVPR 2025 · similarity 0.88)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.88)
- De-Diffusion Makes Text a Strong Cross-Modal Interface (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
