Moyou Android QWen is a fine-tuned and merged model based on Qwen-Image. By utilizing distributed multi-stage training, merging, and difference extraction, the model achieves a clean coexistence of multiple styles, categories, and concepts without mutual contamination. It uses Qwen2.5-VL for captioning to achieve full alignment with CLIP, and boasts strong compatibility with native Qwen-Image LoRAs.

If you care more about whether a model is purely fine-tuned or merged, you may choose other models.

If you prioritize final quality and stability, the Moyou series is definitely your best choice.

For today's high-parameter models, full fine-tuning is no longer suitable for individuals or small studios. Especially for models like Qwen that natively support complex Chinese text, full fine-tuning can severely degrade its originally sound text structure.

Model Features

• First Qwen model to support direct 2K resolution output.
• Excellent portrait generation capability: Freely responds to prompts featuring men, women, young, old, single, or multiple subjects. Depth of field effects (strong, subtle, or none) can also be freely adjusted.
• Broad applicability: Covers various textures (film, studio, web photos, etc.), styles (modern, ancient Chinese, fantasy, cosplay, etc.), and special compositions (grid layouts, popping out of the frame, etc.), enabling stable batch generation for both Chinese and English posters.
• Easy to use: Works directly with Chinese descriptions, or you can use Qwen for prompt interrogation. Highly responsive.

Local Usage Recommendations: Since Qwen-Image requires significant VRAM locally (around 24GB), it is recommended to use it online by selecting "Browse Templates" from the ComfyUI drop-down menu and using the Qwen-Image template workflow.

Please use long Chinese prompts to achieve better results.

• Sampler: euler, res_multistep
• Scheduler: simple, sgm_uniform
• Steps: 20, 30, 50
• CFG: 3.5, 4
• Model Sampling Algorithm AuraFlow Shift: 3.1
• Recommended Resolutions: 1152×1536, 992×1776, 928×1984, 1536×2048 (Both portrait and landscape orientations work for all the above)

For prompt interrogation, it is recommended to use the Ollama node with the qwen3vl model.

If you like this model, please share your generated images to support us. Thank you!