Research map / Image generation / Text-to-image diffusion
UniCombine: Unified Multi-Conditional Combination with Diffusion Transformer
Haoxuan Wang, Jinlong Peng, Qingdong He, Hao Yang, Ying Jin, Jiafu Wu, Xiaobin Hu, Yanjie Pan, Zhenye Gan, Mingmin Chi, Bo Peng, Yabiao Wang
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 Conditional MMDiT Attention (replacing full MMDiT attention) (ablation) | UniCombine training-free · Subject-Insertion | DINO: 88.42 -> 89.01 | Improved |
| add CMMDiT Attention (ablation) | UniCombine training-free · Subject-Insertion | attention operations: 732.17M -> 612.63M | Improved |
| add CMMDiT Attention (ablation) | UniCombine training-free · Subject-Canny | DINO: 86.31 -> 86.88 | Improved |
| add CMMDiT Attention (ablation) | UniCombine training-free · Subject-Depth | MSE: 547.63 -> 507.40 | Improved |
| add CMMDiT Attention (ablation) | UniCombine training-free · Multi-Spatial | MSE: 524.04 -> 519.53 | Improved |
| put trainable LoRA in denoising branch instead of text branch (ablation) | UniCombine training-based · Subject-Insertion | DINO: 92.32 -> 92.96 | Improved |
| add Denoising-LoRA to single-stream blocks as well as dual-stream blocks (ablation) | UniCombine training-based · Subject-Insertion | DINO: 92.38 -> 92.96 | Improved |
| train Denoising-LoRA on multi-conditional data (training-based vs training-free) | UniCombine · Subject-Insertion | FID: 6.37 -> 4.55 | Improved |
| train Denoising-LoRA | UniCombine · Multi-Spatial | MSE: 519.53 -> 165.90 | Improved |
| train Denoising-LoRA | UniCombine · Subject-Canny | CLIP-T: 33.21 -> 33.30 | No clear effect |
| train Denoising-LoRA | UniCombine · Multi-Spatial | CLIP-T: 33.70 -> 33.45 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- Ctrl-X — Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024) · better on 4 of 4 shared results
- ObjectStitch — ObjectStitch: Object Compositing With Diffusion Model (CVPR 2023) · better on 2 of 2 shared results
- UniControlNet — Uni-ControlNet: All-in-One Control to Text-to-Image Diffusion Models (NeurIPS 2023) · better on 2 of 2 shared results
- UniControl — UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- UNIC-Adapter: Unified Image-instruction Adapter with Multi-modal Transformer for Image Generation (CVPR 2025 · similarity 0.89)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.88)
- CoDi: Conditional Diffusion Distillation for Higher-Fidelity and Faster Image Generation (CVPR 2024 · similarity 0.88)
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.87)
- Referee Can Play: An Alternative Approach to Conditional Generation via Model Inversion (ICML 2024 · similarity 0.87)
- ACE: All-round Creator and Editor Following Instructions via Diffusion Transformer (ICLR 2025 · similarity 0.87)
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.
