I made a workflow that automatically inpaints to fix lora bleed when using attention couple. Relies on SAM which isn't the most reliable but it works decently. Configured for 3 people but can add/remove as needed. The refiner(?) for the 2nd region is set up to detect face first, which seems more robust, but only gets the face. Useful if the subjects are close together.
https://files.catbox.moe/ryfm13.png>>76927488Cute!!