Qwen-Image-2.1-viggle-turbo

Note: All model files below must be downloaded via the LM Downloader client. If you do not have the client, Download >>>

This model is released under the strict Qwen Research License, which restricts usage to academic and non-commercial research purposes only. Qwen-Image-2.1-viggle-turbo is a turbo-accelerated variant of Alibaba's open-source Qwen-Image-2.1, created by the Viggle team using DMD (Distilled Matching Distillation) technology to significantly speed up inference. While preserving the original model's text-to-image generation and image editing capabilities, it reduces the number of required sampling steps from 40 to just 4, achieving approximately 6x faster inference without relying on classifier-free guidance (CFG) for stable output. Provided as a LoRA adapter, it can be directly loaded on top of the base Qwen-Image-2.1 model and is compatible with mainstream inference frameworks such as Diffusers, supporting both text-to-image generation and image editing at resolutions up to 1K. It is important to note that this model supports a maximum of only 1 to 3 reference images for editing tasks, and attempting to input more than three reference images will result in degraded or failed performance, so users working with multi-reference scenarios should plan their input strategy accordingly.