Need help with a reliable multi-character workflow for ComfyUI / SDXL

Hi everyone,

Seeing @fateno4455131’s “Hololive Bounty ! Drawn to Dawn Celebration !!! (47500 buzz)” made me revisit a problem that has been bothering me for a long time: multi-character feature bleeding.

I can get good results when I generate each character on their own, but when I try to put both of them into the same image, the model starts mixing their key traits.

For example:

• one character is supposed to have a halo, tentacles, and a purple eldritch vibe

• the other is supposed to have phoenix wings, fire, and an orange/teal vibe

Separately, both images look great.
Together, the model often mixes halos, wings, colors, and effects between them.

Here are my examples:

Character A

Character B

Combined attempt

The combined image gets close to what I want composition-wise, but the identity separation is still unreliable.

I’ve already tried:

• BREAK-based prompt separation

• left/right emphasis

• stronger weighting

• positive/negative prompt control

But I still don’t have a workflow I can really trust for this type of two-character scene.

So I wanted to ask:

What is the best current workflow in ComfyUI / SDXL for keeping two visually distinct characters separated in one image?

If anyone has good results with:

• Regional Prompting

• Latent Couple

• Conditioning Set Area

• IPAdapter

• ControlNet

• layered workflows

• inpaint-based correction

please let me know.

Thanks.