Random Conditioning with Distillation for Data-Efficient Diffusion Model Compression
Dohyun Kim, Sehwan Park, Geonhee Han, Seung Wook Kim, Paul Hongsuck Seo · 2025
Read Random Conditioning with Distillation for Data-Efficient Diffusion Model Compression by Dohyun Kim, Sehwan Park, Geonhee Han, Seung Wook Kim, Paul Hongsuck Seo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-02011.txt.
Description
Shrinking a giant image generator into a nimble student model usually needs piles of training images. Dohyun Kim and co-authors propose Random Conditioning in 2025, a distillation trick that pairs noised images with randomly chosen text prompts instead of matched ones. Freed from condition-specific pictures, the student explores the whole prompt space while copying its teacher.
The student even learns to render concepts absent from its training images, borrowed through the teacher's responses to random prompts. Applied to conditional diffusion models, the method improved both quality and efficiency. The authors release code and models, aiming at cheaper, wider deployment of generative tools.
Catalog record
| CREATOR | Dohyun Kim, Sehwan Park, Geonhee Han, Seung Wook Kim, Paul Hongsuck Seo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-02011