AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

RAPID$^3$: Tri-Level Reinforced Acceleration Policies for Diffusion Transformer

Wangbo Zhao, Yizeng Han, Zhiwei Tang, Jiasheng Tang, Pengfei Zhou, Kai Wang, Bohan Zhuang, Zhangyang Wang, Fan Wang, Yang You

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration

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 learned per-image tri-level policies instead of static or hand-designed accelerationSD3 · COCO, HPS, GenEvalspeedup / quality: 2.92x speed; GenEval overall 63.48 vs 61.67 (SD3 9-steps)Improved
use tri-level policies instead of step-only dynamic policy (TPDM)SD3 · GenEval OverallOverall: 60.70 -> 63.48Improved
use RAPID^3 instead of fine-tuned DyFLUXFLUX · FLUX latency and training costlatency (s): 13.93 -> 8.30; GPU hours 38,000 -> 400Improved
add cache-reuse policy to step-skip (ablation)RAPID^3 · HPSHPS score: 27.75 -> 27.86Improved
add sparse-attention policy to step+cache (ablation)RAPID^3 · HPSHPS score: 27.86 -> 28.07Improved
add step-skip policy to cache+sparse (ablation)RAPID^3 · COCOAesthetic score: 5.13 -> 5.26Improved
add discriminator reward to ImageReward reward (ablation)RAPID^3 · HPSHPS score: 27.72 -> 28.07Improved
add discriminator reward to ImageReward reward (ablation)RAPID^3 · ImageRewardImageReward: 0.9605 -> 0.9574Hurt
use discriminator reward alone without ImageReward (ablation)RAPID^3 · HPSHPS score: 28.07 -> 27.96Hurt
replace manual combination of step, cache and sparse acceleration with learned policies (ablation)SD3 · COCO CLIPCLIP score: 31.43 (manual-1) -> 32.09Improved
replace GRPO with RLOO (ablation)RAPID^3 · COCO, HPSHPS score: 28.07 -> 28.06No clear effect
increase text-only training prompts from 20K to 40K (ablation)RAPID^3 · HPSHPS score: 28.07 -> 28.26Improved

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