Research map / Image generation / Fast diffusion sampling
HiGS: History-Guided Sampling for Plug-and-Play Enhancement of Diffusion Models
Seyedmorteza Sadat, Farnood Salehi, Romann M. Weber
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 HiGS history guidance to the sampler | SiT-XL + REPA · ImageNet 256x256 unguided | FID: 1.83 -> 1.61 | Improved |
| add HiGS to guided sampling with 40 steps instead of 250 | SiT-XL + REPA and DiT-XL/2 · ImageNet 256x256 with CFG | FID | Improved |
| add HiGS guidance | Stable Diffusion 3 · sampling speed | iterations per second: about 6.50 with and without | No clear effect |
| add HiGS to CLIP score evaluation | Stable Diffusion models · text-to-image | CLIP score | No clear effect |
| use CFG-guided predictions instead of conditional predictions as history buffer (ablation) | HiGS · text-to-image | quality | Improved |
| add DCT high-pass frequency filtering to the update term (ablation) | HiGS · text-to-image | color artifacts | Improved |
| add orthogonal projection of the update (ablation) | HiGS · text-to-image | oversaturation | Improved |
| increase HiGS scale beyond about 3 (ablation) | HiGS · multiple models | quality | Hurt |
| replace square-root weight schedule with constant or linear schedule (ablation) | HiGS · multiple models | quality metrics | No clear effect |
| replace EMA history function with other averages (ablation) | HiGS · multiple models | quality metrics | No clear effect |
| lower t_min or reduce t_max in the weight schedule (ablation) | HiGS · multiple models | quality | Hurt |
| add HiGS on top of distilled diffusion models and APG | distilled models and APG · text-to-image | HPS win rate | Improved |
Closest papers by meaning
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- Plug-and-Play Diffusion Distillation (CVPR 2024 · similarity 0.89)
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · similarity 0.89)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.89)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (NeurIPS 2025 · similarity 0.89)
- CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling (ICLR 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.
