Research map / Image generation / Fast diffusion sampling
Manifold-Optimal Guidance: A Unified Riemannian Control View of Diffusion Guidance
Zexi Jia, Pengcheng Luo, Zhengyao Fang, Jinchao Zhang, Jie Zhou
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 |
|---|---|---|---|
| replace Euclidean CFG with Auto-MOG | SD-XL · COCO 512 (w=15) | FID: 22.29 -> 21.60 | Improved |
| replace CFG with Auto-MOG | DiT-XL/2 · ImageNet 256 | FID: 19.14 -> 8.78 | Improved |
| replace CFG with Auto-MOG | SD-XL · COCO 512 (w=15) | Saturation: 0.28 -> 0.17 | Improved |
| add adaptive energy-balanced beta (Auto-MOG) over fixed-beta MOG-Score (ablation) | MOG-Score · SD-XL ablation | FID: 25.75 -> 25.10 | Improved |
| add feature-diagonal metric to score-based metric (ablation) | MOG-Score · SD-XL ablation | FID: 25.75 -> 25.45 | Improved |
| add feature-diagonal metric to score-based metric (ablation) | MOG-Score · SD-XL ablation | latency: 1.01x -> 1.08x | Hurt |
| increase anisotropy ratio from 1 to 10 (ablation) | MOG-Score · SD-XL | FID: 26.20 -> 25.75 | Improved |
| increase anisotropy ratio from 10 to 50 (ablation) | MOG-Score · SD-XL | CLIP score: 34.35 -> 32.10 | Hurt |
| vary rho between 5 and 20 (ablation) | MOG-Score · SD-XL | FID: 25.85 / 25.75 / 25.80 | No clear effect |
| use Auto-MOG instead of best-tuned CFG (w=7.5) (ablation) | CFG · SD-XL | FID: 25.90 -> 25.10 | Improved |
| use Auto-MOG instead of linear guidance decay (ablation) | CFG linear schedule · SD-XL | FID: 25.80 -> 25.10 | Improved |
| increase CFG scale from 5 to 15 (ablation) | CFG · SD-XL | FID: 26.05 -> 26.29 | Hurt |
Closest papers by meaning
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.87)
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · similarity 0.87)
- Feedback Guidance of Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale (ICML 2024 · similarity 0.86)
- Training-Free Guidance Beyond Differentiability: Scalable Path Steering with Tree Search in Diffusion and Flow Models (NeurIPS 2025 · similarity 0.86)
- Diffusion Models without Classifier-free Guidance (NeurIPS 2026 · similarity 0.86)
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.
