AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

WaDi: Weight Direction-aware Distillation for One-step Image Synthesis

Lei Wang, Yang Cheng, Senmao Li, Ge Wu, Yaxing Wang, Jian Yang

CVPR 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace one-step model's weight direction with the multi-step teacher's (ablation)DMD2 (SD 1.5) · COCO FIDFID: +241.3 FID, -0.18 CLIPHurt
replace one-step model's weight norm with the teacher's (ablation)DMD2 (SD 1.5) · COCO FIDFID: +0.7 FID, unchanged CLIPNo clear effect
use LoRaD low-rank rotation adapter instead of LoRA (ablation)WaDi (SD 1.5, VSD loss) · COCO 2017FID: 25.27 -> 20.86Improved
use LoRaD instead of DoRA (ablation)WaDi (SD 1.5, VSD loss) · COCO 2017FID: 26.56 -> 20.86Improved
use LoRaD instead of full fine-tuning (ablation)WaDi (SD 1.5, VSD loss) · COCO 2017FID: 23.30 -> 20.86Improved
freeze DoRA norm (ablation)DoRA · COCO 2017FID: 26.56 -> 24.52Improved
increase student LoRaD rank from 64 to 256 (ablation)WaDi (SD 1.5) · COCO 2014FID: 13.64 -> 10.79Improved
increase student LoRaD rank from 256 to 512 (ablation)WaDi (SD 1.5) · COCO 2014FID: 10.79 -> 12.75Hurt
lower fake-model rank from 32 to 16 (ablation)WaDi (SD 1.5) · COCO 2014FID: 10.79 -> 17.53Hurt
raise fake-model rank from 32 to 64 (ablation)WaDi (SD 1.5) · COCO 2014FID: 10.79 -> 16.98Hurt
distill with WaDi instead of DMD2 reproductionSD 1.5 one-step · COCO 2014FID: 12.96 -> 10.79Improved
apply WaDi to ControlNetControlNet · inference timeseconds: 4.73s -> 0.65sImproved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy