Kissing Dolls -- Multistep Regional Couple Workflow

If you have two dolls, what is more human than making them kiss?

This is a simple workflow for generating two distinct characters interacting, using regional prompting to reduce prompt bleed. It works as follows:

1. A layout image is generated using a model that is good at composition

2. CLIPSeg is used to identify the characters by a descriptor (e.g. man, woman)

3. Regional conditioning is applied based on a mask

4. Image-2-image is used with a second model

5. Upscale and cleanup using the second model

Because of the use of CLIPSeg it performs better with opposite-sex characters where gender can be used as the descriptor. However images can be very successfully generated using alternative descriptors with some thought and patience.

As with my previous workflows, none of this is particularly original or groundbreaking, but I've tried to present it in a tidy and understandable way in the hope that it will be helpful to others!