upsampling step

Generate high-quality images 5x faster with our new 6-step LoRA

Generating high-quality images doesn't have to mean waiting through dozens of steps. Our new LoRA adapters for Qwen-Image-2.1 cut the process to just six steps, saving 41 to 80 seconds per generation. The upsampled variant even removes the separate prompt-upsampling step entirely, streamlining the pipeline. This distillation approach keeps your workflow unchanged while dramatically reducing diffusion cost. For more on how open-source communities are reshaping access to AI research, explore our coverage of Papers with Code's new open-source release.