Set negative prompts in captions ?

[Zokreb]

Hello everyone !

First of all, big thanks to you Bernard for your incredible work, the time involved, and your dedication.
Your GUI for the Kohya scripts is impressive and very useful and all of that for free!

Now to all the savvy community, I haven't been able to find an answer in the readme's (maybe I don't look well enough):

Now, to the question:

How could I input negative prompts in addition to my prompts in training captions. ?

What I would like to try is to counter the effect in one prompt in my training data with a negative counter part in my regularization images and try to look if possible to transform one specific style to another.

Let's say I only have sketches of a character or 3d renders, and I have lot's of real life photography of the regularization images, would it be possible to have prompts looking like this :

#training images :
sketch of robotdog dog --n photo

#regularization images :
photo of dog --n sketch

with robotdog being the thing i want to train on and --n the command switch to set a negative token...

in that sense, the training process would recognize both the differences between the characteristics of the robotdog vs a dog, but would also be able to generalize the characteristics of a sketch and a photograph ? Because otherwise, the robotdog prompt would systematically lead to drawings, right ?

English is not my native language, sorry if this is not clear.

Or is there a simpler way to do this ?

Thank you.

[Zokreb]

I suppose latest update with support for weighted captions kind of answers this question.
Thank you for the updates :)