Research map / Image generation / Text-to-image diffusion
DivControl: Knowledge Diversion for Controllable Image Generation
Yucheng Xie, Fu Feng, Ruixiao Shi, Jing Wang, Yong Rui, Xin Geng
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 knowledge diversion (SVD learngenes and tailors with dynamic gate) (ablation) | ControlNet on PixArt-δ · 8 basic conditions (average) | LPIPS / SSIM / CLIP-I: LPIPS 0.328 -> 0.302; SSIM 0.663 -> 0.693; CLIP-I 89.43 -> 90.63 | Improved |
| add REPA on top of knowledge diversion (ablation) | DivControl · 8 basic conditions (average) | LPIPS / SSIM / CLIP-I: LPIPS 0.302 -> 0.290; SSIM 0.693 -> 0.696; CLIP-I 90.63 -> 91.31 | Improved |
| use alignment weight 0.05 instead of 0.5 (ablation) | REPA · basic conditions | FID: 40.76 -> 39.84 | Improved |
| align at layer 4 instead of layer 2 or 6 (ablation) | REPA · basic conditions | LPIPS: 0.301 -> 0.290 (layer 2); 0.303 (layer 6) | Improved |
| DivControl vs CtrLoRA-SD training cost | DivControl · training cost | GPU hours: 6000 -> 165 | Improved |
| DivControl vs CtrLoRA-PA on basic conditions | DivControl · 8 basic conditions (average) | CLIP-I: 91.85 -> 92.29 | Improved |
| DivControl vs CtrLoRA-SD on average CLIP-I | DivControl · 8 basic conditions (average) | CLIP-I: 92.24 -> 92.29 | No clear effect |
| finetune tailors only on 200 images for high-shift novel conditions | DivControl · novel conditions (average) | CLIP-I: 92.34 -> 93.07 vs CtrLoRA-PA | Improved |
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.
- UniControl — UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild (NeurIPS 2023) · better on 4 of 4 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
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.91)
- Universal Few-shot Spatial Control for Diffusion Models (NeurIPS 2025 · similarity 0.89)
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.89)
- Simplifying Control Mechanism in Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.89)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.89)
- D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models (NeurIPS 2026 · 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.
