How to Train Multiple Single-Character LoRAs for Direct Multi-Character Image Generation
Simply adding multi-character images for each outfit or variant of the character to the training dataset allows the resulting LoRA to directly render multi-character images—without requiring regional prompt control—when paired with other LoRAs trained using the same method. Testing indicates that adding just two multi-character images per outfit/variant is sufficient to grant the LoRA this capability.
Training multiple single-character LoRAs with this approach is simpler than training a single multi-character LoRA, and it makes adding new characters or outfits far more convenient. However, invoking them during generation can be more cumbersome, so you can choose between training multiple single-character LoRAs or a single multi-character LoRA based on your specific needs.
Thank you for other advices it's really worked and improved multicharacter loras