Models · The Decoder ·
Alibaba's Qwen-Image-2.0 doubles compression and cuts generation steps from 40 to 4
Alibaba published a technical report on Qwen-Image-2.0, an image model that uses stronger compression, a redesigned transformer for training stability, and a module that expands short prompts into more detailed ones. A distilled version can generate in four denoising steps, down from 40, and the model ranks 9th on LMAr