A quantized version of the Qwen-Image-2.1 Uncensored model has been released

Developers presented Qwen-Image-2.1-Uncensored-GGUF — a version of the multimodal neural network adapted for local launch with content filters disabled. The model is optimized to work in the ComfyUI environment, allowing users to generate images on consumer equipment.

To ensure a balance between stability and generation quality, it is recommended to use the Q4_K_M quantization format. The size of this file is 4.6 GB, making the model accessible for launch on video cards with limited video memory. The use of the GGUF format allows for efficient distribution of the load between the GPU and CPU.

The model is based on the Qwen-Image-2.1 architecture, but in this iteration, software restrictions on generating certain types of visual content have been lifted. The release is targeted at enthusiasts engaged in local deployment of neural networks and searching for alternative ways to manage the behavior of generative models.

This release expands the ecosystem of tools for working with open weights from Alibaba, providing users with more control over the image synthesis process. The use of quantized versions in the GGUF format lowers the entry threshold for local experiments with multimodal systems, without requiring server capabilities.