What if you could train AI image generators 4x faster without losing quality?

NVIDIA, HKU, and MIT researchers present Sol-RL.

They use a clever two-stage process: first, a super-fast, low-precision search creates a huge pool of image candidates. Then, they regenerate only the… pic.twitter.com/KokLk4tmr9

— 机器之心 JIQIZHIXIN (@jiqizhixin) 23 avril 2026

What if you could train AI image generators 4x faster without losing quality? NVIDIA, HKU, and MIT researchers present Sol-RL. They use a clever two-stage process: first, a super-fast, low-precision search creates a huge pool of image candidates. Then, they regenerate only the

→ View original post on X — @jiqizhixin,