y0news
← Feed
Back to feed
🧠 AI🟢 BullishImportance 7/10

LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration

arXiv – CS AI|Peiliang Cai, Jiacheng Liu, Haowen Xu, Xinyu Wang, Chang Zou, Linfeng Zhang|
🤖AI Summary

Researchers propose LESA, a new framework that accelerates Diffusion Transformers (DiTs) by up to 6.25x using learnable predictors and Kolmogorov-Arnold Networks. The method achieves significant speedups while maintaining or improving generation quality in text-to-image and text-to-video synthesis tasks.

Key Takeaways
  • LESA framework achieves 5.00x acceleration on FLUX.1-dev with only 1.0% quality drop.
  • The method delivers 6.25x speedup on Qwen-Image with 20.2% quality improvement over previous state-of-the-art.
  • Multi-stage, multi-expert architecture assigns specialized predictors to different noise-level stages for better accuracy.
  • Kolmogorov-Arnold Networks are used to learn temporal feature mappings from data.
  • Framework demonstrates strong generalization across different text-to-image and text-to-video models.
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Connect Wallet to AI →How it works
Related Articles