managing background

[ilune]

I want to generate the image of a woman with a background of a battle, something that should be totally different

logo of a woman with a background BREAK
woman caucasian redhead, medieval, dress ,woman BREAK
 medieval battle with sword and blood pencil style gray gradients ,background

I selected the divide mode «prompt», but it seems not to work, i have the woman with a sword and blood on her ...
With prompt ex, it doesn't work at all.

Any ideas how I can do this ?

[Symbiomatrix]

Are you getting 2 additional outputs?

[Symbiomatrix]

You're not following what I'm saying.
What result are you getting for just medieval battle with sword and blood ?

[ilune]

Hum,

This

medieval battle with sword and blood pencil style gray gradients 
Negative prompt: easynegative
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 1239471019, Size: 512x512, Model hash: e04b020012, Model: rpg_V4, Version: v1.2.1

Gives this

If i change text a little

medieval battle with sword and blood, luis royo, pencil art
Negative prompt: easynegative
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 1754467351, Size: 512x512, Model hash: e04b020012, Model: rpg_V4, Version: v1.2.1

It's better

Without selecting region prompt this

logo of a woman with a background BREAK
woman caucasian redhead, medieval, dress ,woman BREAK 
medieval battle with sword and blood, luis royo, pencil art ,background

Negative prompt: easynegative
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 1777564987, Size: 512x512, Model hash: e04b020012, Model: rpg_V4, Version: v1.2.1

gives this

The same with region prompt

logo of a woman with a background BREAK
woman caucasian redhead, medieval, dress ,woman BREAK 
medieval battle with sword and blood, luis royo, pencil art ,background
Negative prompt: easynegative
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 725000859, Size: 512x512, Model hash: e04b020012, Model: rpg_V4, Version: v1.2.1, RP Active: True, RP Divide mode: Prompt, RP Calc Mode: Attention, RP Ratios: "1,3; 1,1,1,1; 2,1,1,1; 1,1,1,1;1,3", RP Base Ratios: 0.3, RP Use Base: True, RP Use Common: False, RP Use Ncommon: False, RP Change AND: False, RP LoRA Neg Te Ratios: 0, RP LoRA Neg U Ratios: 0, RP threshold: 0.4

gives

[Symbiomatrix]

I think some concept bleed is inevitable. Unlike multi subject render and others, prompt mode seems to train the subprompt alongside the main prompt at every step, so if your scene is a battle you get a warrior woman because that's what the model knows should happen in a battle, I believe.
I'm not entirely sure whether the negs are connected respectively, might be useful to remove violence from the woman subprompt.
"Logo of" is clearly overpowering the scene's style, without it the results are at least passably close.

Adding distance and a non warrior woman (noble / peasant) helps a little.

[ilune]

Thank you for you time, I search a little more and get something using this:

close-up view of Tanaelle, fields in the distance in a a circle frame BREAK
woman caucasian redhead, medieval, dress ,Tanaelle BREAK 
medieval battle of armies in the fields, dark clouds, lightning, luis royo, pencil art ,fields 

I don't use «woman» but a name, i don't speak of battle but of fields. Not perfect but i have some success ^^.

[Symbiomatrix]

Aye, that's pretty much the technique, it seems. Personally I still prefer my regions to be predictable, but it's lightyears ahead of cutoff.