Research map / Image generation / Fast diffusion sampling
On Distillation of Guided Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Distill guided diffusion into a single w-conditioned model and progressively reduce the sampling steps | Classifier-free guided diffusion · LAION 512x512, w=8.0 | FID: 98.9 -> 37.3 | Helped |
| Distill guided diffusion into a single w-conditioned model and progressively reduce the sampling steps | Classifier-free guided diffusion · LAION 512x512, w=8.0 | CLIP: 0.20 -> 0.27 | Helped |
| Use 2-step guided distillation rather than 2-step DDIM sampling | Stable Diffusion Inpainting · COCO2017 inpainting, w=4.0 | FID: 109.35 -> 29.50 | Helped |
| Use 4-step guided distillation rather than 4-step DDIM sampling | Stable Diffusion Inpainting · COCO2017 inpainting, w=4.0 | FID: 26.89 -> 24.90 | Helped |
| Train one model across a guidance interval instead of training with fixed w (ablation) | ImageNet 64x64 distilled model · ImageNet 64x64, w=0.3 | FID: 2.32 vs 2.05 | No clear effect |
| Use stochastic rather than deterministic sampling (ablation) | ImageNet 64x64 distilled model · ImageNet 64x64, w=0.3, 8 steps | IS: 76.01 -> 83.00 | Helped |
| Use stochastic rather than deterministic sampling (ablation) | ImageNet 64x64 distilled model · ImageNet 64x64, w=0.3, 8 steps | FID: 2.05 -> 2.31 | Hurt |
Closest papers by meaning
- Plug-and-Play Diffusion Distillation (CVPR 2024 · similarity 0.95)
- Elucidating the design space of classifier-guided diffusion generation (ICLR 2024 · similarity 0.93)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.92)
- Data-free Distillation of Diffusion Models with Bootstrapping (ICML 2024 · similarity 0.92)
- Fast constrained sampling in pre-trained diffusion models (NeurIPS 2025 · similarity 0.92)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · 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.