The authors froze each pretrained generator and trained a lightweight flow policy to transform standard Gaussian noise before generation.
Steering input noise aligns few-step generators without retraining them
Noise Policy Optimization learns a reusable input-noise transformation that improves black-box rewards while keeping the pretrained generator frozen.
Top University
Woocheol Shin · Taeyoung Yun · Jiwon Sung · Hyeongyu Kang · Jeongmin Park · Jaewoo Lee · +4 more
KAIST · Stanford University · MongooseAI · Carnegie Mellon University · Omelet
Research Digest··3 min read
Shin and colleagues address reward alignment for generators that produce outputs in one or a few evaluations, where conventional reinforcement learning can damage output quality and per-sample search erodes their speed advantage.
Why this paper
From KAIST and 4 others
In one line
Noise Policy Optimization aligns few-step generative models by learning a flow policy over the noise space of a frozen generator.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§