Research map / Image generation / Fast diffusion sampling
SAND: Stage-Aware Noise Decomposition for Training-Free Diffusion Guidance
DaeHyun Kim, Hyo-Jun Lee, Hanul Kim, Yeong Jun Koh
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 |
|---|---|---|---|
| apply SAND guidance | SDXL · COCO unconditional | FID: 126.51 -> 75.25 | Improved |
| apply SAND instead of PAG | SD3 · COCO conditional | FID: 18.13 -> 11.06 | Improved |
| apply PAG | SD3 · COCO unconditional | FID: 101.66 -> 102.74 | No clear effect |
| use decomposed schedule instead of parallel-only noise (ablation) | SAND · SDXL unconditional | FID: 84.31 -> 75.25 | Improved |
| use decomposed noise instead of non-decomposed noise (ablation) | SAND · SDXL unconditional | FID: 76.26 -> 75.25 | Improved |
| use stage-aware schedule instead of swapped schedule (ablation) | SAND · SDXL unconditional | FID: 83.69 -> 75.25 | Improved |
| use stage-aware schedule instead of coupled schedules (ablation) | SAND · SDXL unconditional | FID: 83.18 -> 75.25 | Improved |
| perturb mid layers instead of down or up layers (ablation) | SAND · SDXL unconditional | FID: 119.46 -> 75.25 | Improved |
| apply SAND alone to FLUX.1-dev | SAND · FLUX.1-dev DrawBench | CLIPScore: 26.61 -> 22.65 | Hurt |
| add SAND on top of CFG | FLUX.1-dev · DrawBench | ImageReward: 0.9935 -> 1.004 | Improved |
| apply SAND to Wan2.2 | Wan2.2 · VBench | imaging quality: 0.5975 -> 0.6134 | Improved |
Closest papers by meaning
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.90)
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · similarity 0.90)
- Test-Time Scaling of Diffusion Models via Noise Trajectory Search (NeurIPS 2025 · similarity 0.90)
- One More Step: A Versatile Plug-and-Play Module for Rectifying Diffusion Schedule Flaws and Enhancing Low-Frequency Controls (CVPR 2024 · similarity 0.90)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.89)
- DeepCache: Accelerating Diffusion Models for Free (CVPR 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.
