I made a workflow to be able to do consistent character in different scene, including action pose. Using IP Adapter, i could manage to have a good consistent to the character. It's far from being perfect, but it closed enough.

Some Time you have to cheat it, and use you own pose reference image, sometime you have to play a little with the mask, cause IP doesn't understand it well with interaction.. but somehow, it still work very well :)

Devlopper note :

You better use Green Background for character ref. It work better.. all the token IP adaptor will read will be your character and maybe 1 or 2 token  for the green background.

Green is better then white in my experience.

you have to have a simple promt to help the workflow.. ex for the male character the prompt is : 1 bald man with brown apron.. for 2 character it will be : 1 Bald man with brown apron and 1 blond girl with blue dress.

CLOUD SETUP

There is some model you will have to change to make it work, fallow the line, when it block, just try out some model were it bug... for SD1.5 use Model.psytorch for clipmodel, FacePlus V2 for Face, IP Plus for the model for BG and Character extraction.

Future plan :

I plan to add more character to the scene, I don t know what is the limit..

Note : Your pose picture will guide the latent image to produce your image, so use 512x512 pose picture, or the size you want the result.

V3

Update all the flow, but it does the same thing..

1 Character, 2 Character or 3 Character (Din t test the 3 character)

Think are more automatic

Faster..