AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Importance-Aware Low-Rank Distillation of Diffusion Transformers

Denis Zavadski, Sebastian Heid, Damjan Kalšan, Stefan Roth, Carsten Rother

ECCV 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
use truncated SVD instead of block pruning (single-block probing) (ablation)FLUX.dev blocks · single-block probingmean CMMD per block: 1.18 -> 0.23Improved
use SVDtrunc instead of EcoDiff, training-freeFLUX.dev · 90% paramsavg quality reduction R (%): 14.52 -> 5.24Improved
use SVDtrunc instead of EcoDiff, training-freeFLUX.dev · 80% paramsavg quality reduction R (%): 45.99 -> 6.69Improved
use SVDtrunc instead of EcoDiff, training-freeFLUX.dev · 70% paramsavg quality reduction R (%): 85.50 -> 9.86Improved
use SVDtrunc instead of block pruning, training-freeFLUX.dev · 90% paramsavg quality reduction R (%): 1.00 -> 5.24Hurt
use SVDtrunc instead of block pruning, training-freeFLUX.dev · 70% paramsavg quality reduction R (%): 94.52 -> 9.86Improved
add importance-aware allocation vs random (ablation)SVDtrunc · 57% paramsavg quality reduction R (%): 5.68 -> 2.60Improved
add importance-aware allocation vs uniform (ablation)SVDtrunc · 57% paramsavg quality reduction R (%): 3.19 -> 2.60Improved
add importance-aware allocation vs uniform (ablation)SVDtrunc · 49% paramsavg quality reduction R (%): 6.23 -> 4.68Improved
add importance-aware allocation vs uniform (ablation)SVDtrunc · 68% paramsavg quality reduction R (%): 1.46 -> 0.75Improved
add modular knowledge distillation to rectified-flow fine-tuning (ablation)SVDtrunc · 68% paramsavg quality reduction R (%): 2.12 -> 0.93Improved
use MKD only vs combined objective (ablation)SVDtrunc · 68% paramsavg quality reduction R (%): 0.64 -> 0.93No clear effect

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