6-step Turbo LoRA for Qwen-Image-2.1 released on Hugging Face
linoy_tsaban · x · 2026-10-05
- Developer isHeSatoshi released Qwen-Image-2.1 Turbo v0.2.1 on Hugging Face: a rank-128, 6-step LoRA (648MB) that compresses the base flow-matching model's long probability-flow ODE into a fixed 6-step schedule (sigmas 1.0 to 0.25) with CFG off.
- Training uses distribution matching (GDM, from the DMD family): the frozen Qwen-Image-2.1 acts as teacher; the LoRA student plus a trainable "fake critic" score the same latents, and the score difference approximates the KL gradient between student and data distributions, regressing the student's velocity field toward the teacher's output distribution.
- The repo ships a ComfyUI custom node (turbo.py) and full training details.
More from Multimodal
- 400 images later, a DGX Spark user remains unconvinced by SenseNova U1.5-8B — dh7net · 2026-10-05
- Citrine Photo: free open-source Windows app runs SeedVR2 upscaling without ComfyUI — garionhk · 2026-10-05
- Grok's image generation quality keeps improving, just in time for Halloween — yunta_tsai · 2026-10-05
- GPT Image 2.5 "Sunburst" prompts go viral with analog film-chemistry aesthetics — LudovicCreator · 2026-10-05
- Children's video made with an AI music video app answers "Do Robots Dream?" — Kyrannio · 2026-10-05
- Community-built library of 500 image styles for Krea 2, also works with Kroma and Qwen — Dear-Spend-2865 · 2026-10-05