Qwen Me Anime (Experimental Checkpoint)
Qwen Me Anime is a specialized anime-centric checkpoint trained using the Musubi Trainer tool. This model is designed to capture the vibrant aesthetics of modern anime while maintaining a flexible structure for various character designs.
⚙️ Training Details
Anime Dataset: 2,500 curated anime images.
Regularization Dataset: 3,000 images to maintain model stability.
Status: Experimental. As a work-in-progress, you may encounter occasional imperfections or inconsistencies in certain generations.
💡 Key Characteristics
Anime Stylization: Deeply tuned for sharp lineart and cel-shaded coloring.
Versatility: Capable of handling a wide range of character prompts and backgrounds.
Optimized for Qwen: Built to leverage the unique architecture of the Qwen vision-language foundation.
🚀 Recommended Settings (Important!)
To achieve the best results and overcome the current "experimental" limitations, it is highly recommended to use this checkpoint in combination with a lighting LoRA:
Essential LoRA: Qwen Image Lighting
LoRA Versions: Both 4-steps or 8-steps versions work excellently.
Why? This helps stabilize the output, improve contrast, and clean up visual artifacts.
📝 Usage Suggestions
Prompting: Focus on descriptive anime tags (e.g., masterpiece, high quality, vibrant colors, detailed background).
Sampler: Euler or DPM++ series (Experiment based on your preference).
CFG Scale: 1 - 3 (Depending on your specific LoRA setup).
⚠️ Disclaimer
This model is still in its beta/experimental phase. Feedback and sample images from the community are greatly appreciated to help improve future versions!