Research map / Image generation / Autoregressive and latent generation
OminiControl: Minimal and Universal Control for Diffusion Transformer
Zhenxiong Tan, Songhua Liu, Xingyi Yang, Qiaochu Xue, Xinchao 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 |
|---|---|---|---|
| condition on all transformer blocks instead of early blocks only (ablation) | OminiControl · Canny-to-image | F1 score: 0.23 -> 0.38 | Improved |
| condition on all transformer blocks instead of early blocks only (ablation) | OminiControl · Canny-to-image | FID: 25.66 -> 20.63 | Improved |
| increase LoRA rank from 4 to 16 (ablation) | OminiControl · Canny-to-image | F1 score: 0.380 -> 0.407 | Improved |
| increase LoRA rank from 4 to 16 (ablation) | OminiControl · Canny-to-image | FID: 20.63 -> 19.71 | Improved |
| use LoRA rank 1 instead of 4 (ablation) | OminiControl · Canny-to-image | CLIP score: 0.761 -> 0.765 | No clear effect |
| use unified sequence concatenation instead of direct feature adding (ablation) | OminiControl · Canny-to-image training loss | training loss | Improved |
| shift condition position indices instead of sharing positions (ablation) | OminiControl · subject-driven generation | training convergence and quality | Improved |
| train on Subjects200K instead of data augmentation (ablation) | OminiControl · subject-driven generation | qualitative novelty of views | Improved |
| keep LoRA on normalization layers and W_Q, W_K projections (ablation) | OminiControl · conditional rendering | qualitative control quality | Improved |
| keep LoRA on value projections W_V (ablation) | OminiControl · conditional rendering | qualitative control quality | No clear effect |
| use OminiControl instead of ControlNet Pro on deblurring | OminiControl · Deblur | MSE: 338 -> 62 | Improved |
| use OminiControl instead of ControlNet Pro on colorization | OminiControl · Colorization | MSE: 994 -> 73 | Improved |
Later papers that compared with this method
15 later accepted papers report OminiControl as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.
- DynFusion: Rethinking Condition Fusion for Adaptive Multi-Conditional Text-to-Image Generation (CVPR 2026 · better on 8 of 8 shared results)
- Gated Condition Injection without Multimodal Attention: Towards Controllable Linear-Attention Transformers (CVPR 2026 · better on 5 of 5 shared results)
- Group Editing: Edit Multiple Images in One Go (CVPR 2026 · better on 5 of 5 shared results)
- EchoGen: Generating Visual Echoes in Any Scene via Feed-Forward Subject-Driven Auto-Regressive Model (ICLR 2026 · better on 3 of 3 shared results)
- SIGMA-GEN: STRUCTURE AND IDENTITY GUIDED MULTI-SUBJECT ASSEMBLY FOR IMAGE GENERATION (ICLR 2026 · better on 3 of 3 shared results)
- HierEdit: Region-Aware Hierarchical Diffusion for Efficient High-Resolution Editing (CVPR 2026 · better on 2 of 3 shared results)
- Dual Optimal Transport for Multi-Concept Composition: Structural Alignment and Texture Injection in Diffusion Models (ICML 2026 · better on 1 of 1 shared result)
- PositionIC: Unified Position and Identity Consistency for Image Customization (CVPR 2026 · better on 1 of 1 shared result)
7 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 3 of 5 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
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.87)
- U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers (NeurIPS 2024 · similarity 0.87)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.87)
- Dynamic Diffusion Transformer (ICLR 2025 · similarity 0.87)
- FlexiDiT: Your Diffusion Transformer Can Easily Generate High-Quality Samples with Less Compute (CVPR 2025 · similarity 0.87)
- FreeControl: Efficient, Training-Free Structural Control via One-Step Attention Extraction (NeurIPS 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.
