Z-Image Turbo++ achieves 2-step image generation via teacher-aligned distillation

Arxiv CS.CV (computer vision)

Paper introduces Z-Image Turbo++, a 2-step image generation model distilled from an 8-step teacher. The method uses teacher-aligned end-to-end distillation to maintain high fidelity, overcoming challenges in few-step diffusion.

Text-to-imageDiffusionDistillation

See 1 more source

[Papers] - Z Image Turbo ++ - High-Fidelity 2-Step Image Generation via Teacher-Aligned End-to-End Distillation5 days agoCrazy-Repeat-2006Discuss

Arxiv CS.CV (computer vision)·Dongyang LiuRuoyi DuDavid LiuDengyang JiangLiangchen LiQilong WuZhen LiSteven C. H. HoiHongsheng LiPeng Gao

Jun 12, 4:00 AM