Qwen-Image-Flash: Beyond Objective Design
- Published
- Source
- arXiv
- Paper number
- 358
- Field
- Computer Vision
- arXiv ID
- 2606.03746
Key points
- It trains a 4-NFE student model distilled from Qwen-Image-2.0 using distribution matching distillation, or DMD.
- A counterintuitive data finding emerges: increasing diversity hurts performance, while consistent single-category data transfers better.
- Step-wise multi-teacher guidance combines complementary strengths across teacher models and stabilizes training.
- It achieves the best overall performance at a balanced T2I and Edit ratio, and edit supervision also improves generation quality.
- Qwen-Image-Flash produces high-quality results with only 4 NFE in scenarios such as poster generation.
Paper links
External research summaries. These are not HDATF publications or measured product results.