Research map / Image generation / Fast diffusion sampling
Learn to Guide Your Diffusion Model
Alexandre Galashov, Ashwini Pokle, Arnaud Doucet, Arthur Gretton, Mauricio Delbracio, Valentin De Bortoli
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 |
|---|---|---|---|
| learn conditioning-dependent guidance weights via self-consistency instead of constant guidance | diffusion model · ImageNet 64x64 | FID: 2.40 -> 1.99 | Improved |
| learn guidance weights via self-consistency instead of unguided sampling | diffusion model · ImageNet 64x64 | FID: 4.46 -> 1.99 | Improved |
| learned self-consistency weights instead of limited interval guidance | diffusion model · ImageNet 64x64 | FID: 2.11 -> 1.99 | Improved |
| condition guidance weights on class c in self-consistency (ablation) | self-consistency guidance · ImageNet 64x64 | FID: 2.07 -> 1.99 | Improved |
| condition guidance weights on class c in self-consistency (ablation) | self-consistency guidance · ImageNet 64x64 | Inception Score: 76.7 -> 73.62 | Hurt |
| use l2 objective instead of self-consistency MMD objective (ablation) | learned guidance · ImageNet 64x64 | FID: 1.99 -> 2.09 | Hurt |
| learned self-consistency weights instead of unguided sampling | diffusion model · CelebA 64x64 | FID: 2.44 -> 2.10 | Improved |
| add conditioning to self-consistency guidance weights (ablation) | self-consistency guidance · CelebA 64x64 | FID: 2.28 -> 2.10 | Improved |
| constant guidance w=0.01 instead of none | diffusion model · CelebA 64x64 | FID: 2.44 -> 2.45 | No clear effect |
| learned self-consistency guidance with negative prompt instead of unguided | flow matching T2I model · MS COCO 512x512 | FID: 24.74 -> 18.01 | Improved |
| learned self-consistency guidance instead of constant w=7.5 | flow matching T2I model · MS COCO 512x512 | FID: 31.2 -> 18.01 | Improved |
| learned self-consistency guidance instead of constant w=7.5 | flow matching T2I model · MS COCO 512x512 | CLIP Score: 0.306 -> 0.295 | Hurt |
Closest papers by meaning
- Rectified Diffusion Guidance for Conditional Generation (CVPR 2025 · similarity 0.94)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.94)
- Diffusion Models without Classifier-free Guidance (NeurIPS 2026 · similarity 0.93)
- Feedback Guidance of Diffusion Models (NeurIPS 2025 · similarity 0.93)
- Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models (AAAI 2025 · similarity 0.93)
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.92)
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.
