Dragon Ball Super - Qwen Image
This was my first attempt at training a LoRA for Qwen Image.
The results are promising for the overall anime style, but character faces perform average.
Trigger Word: dragon_ball_super
Coverage
Works Well When
Resolution: Use 1280x720 landscape (dataset was trained only on this resolution).
Prompts: Describe character by name and appearance.
Character Count: Solo characters or maximum 2.
LoRA Strength: 0.8 – 0.9
Gotchas
Other Versions: https://huggingface.co/ksol8/dragon_ball_super_qwen_image/tree/main