The stable diffusion webui will split long prompt into pieces to meet the requirement of the text embedding model. All the pieces are independent and treated equally by the following process.

We don't want out prompt to be split at improper position. For example, a long prompt

..., looking at viewer, 1girl ...

may be split into two parts

..., looking at

viewer, 1girl, ...

which may means

..., looking at some unknown place

there is a viewer that is a girl, ...

The above behaviour may happen silently in the background and ruin our result.

This extension prints the processed result of our prompt so that we can find the problem.

To fix the problem, we can just add some commas so that the prompts stay at the right part.

..., , , looking at viewer, 1girl ...

will be split into

..., , ,

looking at viewer, 1girl, ...

This is what we want.

Another way may be use AND to split prompts, but the behaviour is different.

..., AND looking at viewer, 1girl, ...

will become the sum of

...,

looking at viewer, 1girl, ...

see this for details.