AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Training-free Mixed-Resolution Latent Upsampling for Spatially Accelerated Diffusion Transformers

Wongi Jeong, Kyungryeol Lee, Hoigi Seo, Se Young Chun

CVPR 2026 · Image generation › Autoregressive and latent generation · approach: High-resolution diffusion synthesis

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
add NT-Matching with optimized JSD (0.026 vs 0.040) (ablation)RALU · FLUX 7xImageReward: 0.966 -> 0.999Improved
add NT-Matching with optimized JSD (0.026 vs 0.030) (ablation)RALU · FLUX 7xT2I-CompBench: 0.571 -> 0.633Improved
use VAE decode + Canny edge detection instead of latent gradient kernel (ablation)RALU · FLUX 5xImageReward: 0.996 -> 1.022Improved
use adaptive upsampling ratio instead of fixed ratio r=0.2 (ablation)RALU · FLUX 5xImageReward: 1.015 vs 1.022No clear effect
increase early upsampling ratio from low values toward 0.3 (ablation)RALU · FLUX (ImageReward and CLIP-IQA vs TFLOPs)image qualityMixed
upsample edge regions early instead of upsampling all latents late (Bottleneck Sampling)spatial acceleration · FLUX 5xNIQE: 9.16 -> 6.43Improved
combine RALU with TaylorSeer (W=3)RALU · FLUX 5xspeedup: 5.53x -> 7.28xImproved
combine RALU with Delta-DiTRALU · FLUX 5xImageReward: 1.022 -> 0.827Hurt
combine RALU with ToCaRALU · FLUX 5xCLIP-IQA: 0.700 -> 0.522Hurt
apply RALU to timestep-distilled FLUX.1-schnellFLUX.1-schnell · FLUX.1-schnell 4 stepsspeedup: 11.83x -> 15.91xImproved
apply RALU to timestep-distilled FLUX.1-schnellFLUX.1-schnell · FLUX.1-schnell 4 stepsImageReward: 1.055 -> 0.992Hurt
apply RALU to SD3.5L-TurboSD3.5L-Turbo · SD3.5L-Turbo 4 stepsImageReward: 0.840 -> 0.789Hurt

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