Ported from https://modelscope.cn/models/MoYouuu/MYHuman-QWen/summary?version=20251018021646

MoYou Synthetic Human QWen is a fine-tuned and merged model based on Qwen-Image 2509. Through distributed multi-stage training, merging, and differential extraction, the model achieves a clean coexistence of multiple styles, categories, and concepts without cross-contamination. It uses Qwen2.5-VL for captioning, fully aligning with CLIP, and offers strong compatibility with native LoRAs for Qwen-Image.

Model Features:

• The first Qwen model to support native 2K resolution direct generation.
• Outstanding Portrait Generation Capabilities: Seamlessly handles subjects of any gender or age, both single and multiple characters. Allows free adjustment of strong depth-of-field, weak depth-of-field, and no depth-of-field (bokeh) effects.
• Broad Applicability: Covers various textures (film, studio, web photos, etc.), multiple styles (modern, ancient Chinese, fantasy, cosplay, etc.), and diverse special compositions (grid layouts, out-of-frame effects, etc.). Capable of stable batch image generation for both Chinese and English posters.
• Easy to Use: Simply input Chinese text descriptions, or use Qwen for reverse prompting. Excellent responsiveness.

Local Usage Recommendations: Due to the high local VRAM requirement for Qwen-Image (approx. 24GB), it is recommended to use it online by selecting "Browse Templates" from the ComfyUI dropdown menu and loading the Qwen-Image template workflow.

Please use long Chinese prompts to achieve better results. Samplers: euler, res_multistep Schedulers: 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 (All resolutions above support both portrait and landscape orientations)