Research map / Image generation / Fast diffusion sampling
Error as Signal: Stiffness-Aware Diffusion Sampling via Embedded Runge-Kutta Guidance
Inho Kong, Sojin Lee, Youngjoon Hong, Hyunwoo J. Kim
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 ERK-Guid (w_stiff=2.0) to Heun sampler (ablation) | EDM2 Heun, 32 steps · ImageNet-512 | FD-DINOv2: 90.1 -> 82.8 | Improved |
| increase w_stiff from 2.0 to 2.5 (ablation) | ERK-Guid, 32 steps · ImageNet-512 | FD-DINOv2 / FID: 82.8 -> 84.9 FD-DINOv2; FID 2.74 -> 3.03 | Hurt |
| add ERK-Guid (w_stiff=2.0) to Heun sampler (ablation) | EDM2 Heun, 32 steps · ImageNet-512 | FID: 2.58 -> 2.74 | Hurt |
| add ERK-Guid at 16 steps (ablation) | EDM2 Heun · ImageNet-512 | FD-DINOv2: 97.4 -> 88.9 | Improved |
| add ERK-Guid at 8 steps (ablation) | EDM2 Heun · ImageNet-512 | FID: 7.06 -> 4.91 | Improved |
| add ERK-Guid to CFG | CFG, 32 steps · ImageNet-512 | FD-DINOv2: 88.5 -> 83.9 | Improved |
| add ERK-Guid to Autoguidance | Autoguidance, 32 steps · ImageNet-512 | FD-DINOv2: 50.4 -> 47.6 | Improved |
| add ERK-Guid to CFG at 16 steps | CFG, 16 steps · ImageNet-512 | FID: 3.61 -> 3.20 | Improved |
| add ERK-Guid to Autoguidance at 16 steps | Autoguidance, 16 steps · ImageNet-512 | FID: 2.32 -> 1.93 | Improved |
| add ERK-Guid to DPM-Solver | DPM-Solver, 6 NFEs · ImageNet-64 | FID: 44.83 -> 31.59 | Improved |
| add ERK-Guid to DPM-Solver | DPM-Solver, 6 NFEs · FFHQ-64 | FID: 83.17 -> 49.0 | Improved |
| add ERK-Guid to Heun | Heun, 10 NFEs · ImageNet-64 | FID: 16.46 -> 13.85 | Improved |
Closest papers by meaning
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · similarity 0.91)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.90)
- Token Perturbation Guidance for Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale (ICML 2024 · similarity 0.89)
- Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Diffusion Models without Classifier-free Guidance (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.
