Maybe to process the Nano-Banana generated dataset before fine-tuning, and then also to fix the generated Qwen output?
Maybe to process the Nano-Banana generated dataset before fine-tuning, and then also to fix the generated Qwen output?