Research map / Image generation / Fast diffusion sampling
Spectral Guidance for Flexible and Efficient Control of Diffusion Models
Gabriel Moreira, Manuel Marques, Joao Costeira, Chenyan Xiong
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 |
|---|---|---|---|
| use Spectral Guidance instead of the strongest training-free baseline TFG | DDPM 36M · CIFAR-10 / Labels | accuracy: 52.0 -> 89.4 | Improved |
| use Spectral Guidance instead of TFG | DDPM 36M · CIFAR-10 / Labels | FID: 91.7 -> 70.7 | Improved |
| use Spectral Guidance instead of TFG | DDPM 114M · CelebA-HQ / Gender + Age | accuracy: 75.2 -> 91.5 | Improved |
| use Spectral Guidance instead of LGD | DDPM 114M · CelebA-HQ / Gender + Age | log KID: -5.10 -> -2.98 | Hurt |
| use Spectral Guidance instead of TFG | DDPM 121M · ImageNet / Labels | accuracy: 40.9 -> 41.6 | Improved |
| use Spectral Guidance instead of TFG | DDPM 121M · ImageNet / Labels | FID: 176 -> 183 | Hurt |
| use Spectral Guidance instead of TFG for hair-mask guidance | DDPM 114M · CelebA-HQ / Mask | IoU: 0.78 -> 0.80 | Improved |
| use Spectral Guidance instead of TFG for CLIP text guidance | DDPM 114M · CelebA-HQ / CLIP | VQAScore: 0.62 -> 0.64 | Improved |
| use Spectral Guidance instead of TFG for sampling latency | DDPM 114M · CelebA-HQ computational cost | latency (ms/step): 81.2 -> 21.7 | Improved |
| use Spectral Guidance instead of TFG, including offline training cost | DDPM 114M · CelebA-HQ computational cost | effective time for 10k images (h): 22.5 -> 16.9 | Improved |
| increase spectral rank K (8 to 128 and beyond) (ablation) | Spectral Guidance · CIFAR-10 / Labels | accuracy | Improved |
| increase spectral rank K at fixed guidance strength (ablation) | Spectral Guidance · CIFAR-10 / Labels | FID | Hurt |
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.
- FreeDoM — FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model (ICCV 2023) · better on 1 of 1 shared result
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
- Token Perturbation Guidance for Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Variational Control for Guidance in Diffusion Models (ICML 2025 · similarity 0.90)
- Feedback Guidance of Diffusion Models (NeurIPS 2025 · similarity 0.89)
- REG: Rectified Gradient Guidance for Conditional Diffusion Models (ICML 2025 · similarity 0.89)
- Simple Guidance Mechanisms for Discrete Diffusion Models (ICLR 2025 · similarity 0.89)
- Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale (ICML 2024 · similarity 0.89)
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.
