PrunaAI ships 6-step LoRA adapters without prompt upsampling

PrunaAI published 6-step LoRA adapters on Hugging Face for Qwen-Image-2.1, cutting image generation or editing from 40 steps to 6 while keeping the pipeline unchanged. The 6-steps-upsampled variant is described as the first distillation to drop the separate prompt-upsampling step, saving 41 to 80 seconds end to end.

• User builds six small models cheaply via ML Intern2026-10-08
• Microsoft releases 4B vision-language model Mage-VL-Cowork2026-10-31
• Developer builds Halloween haunted house with NVIDIA Nemotron models2026-10-30
• Community releases Gemma-4-31B-Humanlike-Chat fine-tune2026-10-30
• Community releases Qwen 3.8 Swift 1.5 Flash Heretic v2 uncensored quant2026-10-30