The authors keep a 6B-parameter few-step generator frozen and train a parallel corrector with fewer than 20M parameters.
Velocity corrections improve instruction following in few-step text-to-image generation
A frozen 6B-parameter generator keeps its sampling schedule while an 18.6M-parameter corrector learns residual velocity adjustments at early steps.
Chinese Tech
Yating Gao · Zhenlong Zhu · Wen Zihao · Lenan Zhang · Jiexiong Wang · Yang Yang · +3 more
The University of Osaka · Alibaba Group · Peking University · Huazhong University of Science and Technology
Research Digest··2 min read
The authors propose a lightweight velocity correction framework for few-step text-to-image generators.
Why this paper
From Alibaba Group and 3 others
In one line
A sub-20M-parameter velocity residual corrector improves few-step text-to-image instruction following by 17.60 percentage points without altering the frozen generator or adding forward passes.
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)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§