This LoRA enhances Qwen-Image's support for Chinese, English, Korean, Japanese, Spanish, French, Portuguese, Russian, German, and Italian, making the generation effects of prompts more consistent.
It's trained on the Qwen-Image-Self-Generated-Dataset, which includes 1.5 million annotated multilingual prompts, with training code based on DiffSynth-Studio.
The model is mirrored here for convenience.