Chatr Labs today announced the public release of Qwen-Image-2.1 Uncensored Omni, the research group’s most capable unrestricted Qwen-Image-2.1 model to date.
Trained on approximately one million images, Omni uses a single model for both text-to-image generation and natural-language image editing. The release includes nine optimized builds, giving creators and researchers options ranging from high-fidelity BF16 and Q8_0 checkpoints to lower-memory formats designed for local GPU workflows.
Instead of requiring separate models for generation and editing, Omni lets users generate an original image and continue transforming it through natural-language instructions. The model was developed for greater flexibility across character poses, complex compositions, scene changes, outfit modifications and other challenging visual workflows.
The model weights and documentation are available on Hugging Face. Code, technical documentation and inference resources are available on GitHub.
One Model for the Complete Visual Workflow
Qwen-Image-2.1 Uncensored Omni supports two primary workflows through the same model:
- Creating new images from natural-language prompts
- Editing existing images through natural-language instructions
Users can begin with a text prompt, generate an image and then continue refining the result without loading a separate editing model. Editing instructions can be used to modify outfits, environments, colors, lighting, poses and other visual elements while preserving the broader composition.
“Our goal was to build one model that creators could use throughout the entire visual workflow—from the first prompt to the final edit,” said a Chatr Labs spokesperson. “By releasing nine different builds, we are also making that workflow available across a much wider range of local hardware.”
The release is based on a rank-128 Omni adapter trained by Chatr Labs and merged into the Qwen-Image-2.1 transformer. Unlike a few-step distilled model, Omni is intended for full-quality generation and editing workflows where image quality, prompt interpretation and compositional flexibility are prioritized.
Nine Builds for Local AI Creation
Qwen-Image-2.1 Uncensored Omni is available in six GGUF builds and three native ComfyUI-compatible Safetensors builds:
- BF16 GGUF
- Q8_0 GGUF
- Q6_K GGUF
- Q5_K_M GGUF
- Q4_K_M GGUF
- Q4_0 GGUF
- FP8 Safetensors
- INT8 ConvRot Safetensors
- NVFP4 Safetensors
The different builds are intended to help users select the appropriate balance of image quality, model size, performance and GPU memory consumption.
Chatr Labs recommends:
- Q8_0 for results closest to the BF16 model
- Q5_K_M for a strong balance between image quality and file size
- INT8 ConvRot for users seeking high quality in a native ComfyUI format
- NVFP4 for lower-memory workflows on supported newer NVIDIA hardware
- BF16 for users who want the highest-fidelity reference build and have sufficient GPU memory
Recommended starting settings include 35 sampling steps, CFG 5.0 and an output resolution of approximately one megapixel. Detailed prompts that describe composition, subject positioning, lighting, environment and visual style are recommended for more precise control.
Greater Flexibility for Complex Images
Omni was developed to address creative scenarios that can be difficult for general-purpose image models, including diverse character poses, complex body positioning, multi-element compositions and substantial changes to existing images.
The model is intended for experimental creative workflows such as character design, concept development, visual storytelling, scene transformation and iterative image editing.
Qwen-Image-2.1 Uncensored Omni can be used for both original image creation and image-conditioned editing, allowing users to maintain a more consistent workflow across multiple stages of a project.
“Creators should not have to choose between flexibility and accessibility,” the Chatr Labs spokesperson added. “Omni combines generation and editing in one model, while the quantized releases make it possible for more people to explore the model on their own hardware.”
An Unrestricted Research Release
In this release, “uncensored” means that Chatr Labs has not added a prompt filter or output safety checker to the provided inference workflow.
The term does not guarantee that the model will produce every requested result. Outputs continue to depend on the underlying Qwen-Image-2.1 architecture, the prompt, inference settings, quantization format and deployment environment.
Users are responsible for complying with applicable laws, platform policies and license requirements when downloading, deploying or using the model.
The release is distributed under the applicable Qwen Research License Agreement. Use is limited to non-commercial research and evaluation unless separately authorized by the upstream rights holder.
Qwen-Image-2.1 Uncensored Omni is an independent community fine-tune developed by Chatr Labs. Chatr Labs is not affiliated with or endorsed by the Qwen team.
Public Availability
Qwen-Image-2.1 Uncensored Omni is available now through the following official channels:
- Model weights and documentation: Hugging Face
- Code and technical resources: GitHub
- Official website: Chatr Labs
About Chatr Labs
Chatr Labs is an independent generative AI research group focused on openly available image-generation and image-editing technology.
The group develops experimental fine-tunes, model adapters, quantized checkpoints and inference resources intended to make advanced generative media research more accessible to creators, developers and researchers.
Chatr Labs explores models that support flexible creative workflows, local inference and practical experimentation across a range of hardware configurations.
For more information, visit https://chatrlabs.com.