Lightx2v releases Qwen-Image-2.1 Lightning: 4-step distilled LoRA trained on Ascend 910B
LysandreJik · x · 2026-10-14
Lightx2v released Qwen-Image-2.1 Lightning, a 4-step distilled LoRA for Qwen-Image-2.1 supporting text-to-image and image editing, trained with DMD entirely on 16 Huawei Ascend 910B NPUs. The HF model card provides Diffusers BF16 inference examples using a dynamic exponential-shifting noise schedule (4 steps, sigmas [1.0, 0.75, 0.5, 0.25], truecfgscale=1.0, no KV cache); training keeps latents in FP32 so BF16 results may differ. Weights and a demo Space are open-sourced.
Related event: LightX2V Open-Sources 4-Step Distilled LoRA for Qwen-Image-2.1(2 posts)→
More from Multimodal
- MiniMax-generated horror series goes viral: the toy moves every time you blink — fa6637 · 2026-10-16
- AI-generated series "The Dining Car" Ep 4: the world ended, dinner service didn't — Acceptable-Day5621 · 2026-10-16
- Midjourney trick: layer a --profile moodboard over structured prompts for cinematic dark looks — LudovicCreator · 2026-10-16
- Multigen launches: a frontend for GPT-Image to build consistent image collections — Markasp · 2026-10-16
- One-prompt Skill generates an 18-second glitch-style product launch video — yihui_indie · 2026-10-16
- FLUX 3 vs Krea 2 vs Z-Image Turbo vs Qwen 2.1: Four-Way Image Model Comparison — Ok_Contribution8157 · 2026-10-16