Alibaba's Qwen-Image-2.0 doubles compression and cuts generation steps from 40 to 4
Alibaba releases Qwen-Image-2.0, an image generation model with 2x compression, a reworked transformer, automatic prompt expansion, and a distilled variant requiring only 4 denoising steps.
Alibaba's technical report on Qwen-Image-2.0 breaks down how the image model compresses images twice as aggressively as most competitors, stabilizes training with a reworked transformer, and uses a dedicated module that automatically expands short user input into detailed prompts. A distilled version needs just four denoising steps instead of 40.
On LMArena, a platform where users run blind comparisons, Qwen-Image-2.0 currently ranks 9th. The article Alibaba's Qwen-Image-2.0 doubles compression and cuts generation steps from 40 to 4 appeared first on The Decoder.