The Unrestricted Qwen-Image-2.1 Release Built for Creators

Our most capable unrestricted Qwen-Image-2.1 release yet. One model, nine precision options, and maximum creative freedom—from near-BF16 fidelity to a 4.2 GB low-VRAM build.

Generate and edit with the same model, choose the precision that fits your GPU, and keep full local control—without a prompt filter or output safety checker added by this release.

Qwen-Image-2.1 Uncensored Omni Quantized

Chatr Labs Uncensored Omni Quantized is the rank-128 Omni adapter merged into the Qwen-Image-2.1 transformer at strength 1.0 and released as transformer-only quantized checkpoints. The same model supports both text-to-image generation and instruction-based image editing.

Available formats

• GGUF: BF16, Q8_0, Q6_K, Q5_K_M, Q4_K_M, and Q4_0

• Native ComfyUI: FP8, INT8 ConvRot, and NVFP4

GGUF files require the ComfyUI-GGUF custom node. The -comfy.safetensors files use ComfyUI's built-in Load Diffusion Model node.

Recommended settings

• Steps: 35

• CFG: 5.0

• Sampler / scheduler: er_sde / simple

• Negative prompt: a single space

• Resolution: 768×1152 or 1024×1024

This download contains the transformer only. The Qwen-Image-2.1 text encoder and VAE are also required.

The inference package does not add a prompt filter or output safety checker. Results depend on the base model, prompts, and deployment environment. Use must follow Civitai's content policies and the Qwen Research License.

GPU compute used for development and evaluation was provided by Chatr AI.

Model weights & documentation: Hugging Face — Model Weights & Documentation