← sources

lightx2v/Qwen-Image-2.1-Lightning - 4 Step Turbo LoRA

source: reddit (reddit:StableDiffusion)published 2026-10-13 16:53ingested 2026-10-14 02:40

retained text

"A **4-step distilled LoRA for** [**Qwen-Image-2.1**](https://huggingface.co/Qwen/Qwen-Image-2.1), supporting text-to-image generation and image editing. Training was performed entirely on **16 Huawei Ascend 910B NPUs**, using DMD in [LightX2V](https://github.com/ModelTC/LightX2V)."

I think it's rank 16. Pretty small for a turbo lora, but we'll see I guess.

document fetched 2026-10-14 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1x51znr/lightx2vqwenimage21lightning_4_step_turbo_lora