Prompt Separation Issue with [A|B] Syntax in DTG

[namemechan]

I often use the syntax [A_artist|B_artist] for combining art styles. However, when I adapted this to the dtg prompt syntax and placed it in the second line for execution, I encountered an issue where the artist prompt was separated and applied. As a temporary workaround, if I create a wildcard solely for this prompt combination and input it, it functions normally.

[KohakuBlueleaf]

All these sd-webui prompt syntax will result in error or unexpected behaviour since I cannot find a better way to get the prompt after all these syntax parsing
still investigating

[KohakuBlueleaf]

@namemechan Want to confirm
Which one do you want to put into DTG to gen the tags
Since this syntax means you switch from A to B or back in each step

[namemechan]

[https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#alternating-words] This is the method described in the official documentation. If input it like [A_artist|B_artist], the first step applies A_artist and the second step applies B_artist, and this is repeated. You can also mix more prompts like [A_artist|B_artist|C_artist].

Also, I found an additional problem. After creating with DTG, if proceed with hires through the [✨] button, DTG is run again to create a new seed and a new prompt to apply. This causes little distortion at a low denoise strength of about 0.1, so it’s not noticeable, but as the value increases, the picture is greatly distorted.

[KohakuBlueleaf]

[https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#alternating-words] This is the method described in the official documentation. If input it like [A_artist|B_artist], the first step applies A_artist and the second step applies B_artist, and this is repeated. You can also mix more prompts like [A_artist|B_artist|C_artist].

Also, I found an additional problem. After creating with DTG, if proceed with hires through the [✨] button, DTG is run again to create a new seed and a new prompt to apply. This causes little distortion at a low denoise strength of about 0.1, so it’s not noticeable, but as the value increases, the picture is greatly distorted.

I think you totally misunderstand what I want to ask
I know exactly what the syntax doing I even maintained some source code related to it before.
The problem is this is not "mixed" but "switch each step" (prompt_parser.py Line28)
And the problem is DTG will generate totally different prompt if I put each prompt seperatedly

[KohakuBlueleaf]

[https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#alternating-words] This is the method described in the official documentation. If input it like [A_artist|B_artist], the first step applies A_artist and the second step applies B_artist, and this is repeated. You can also mix more prompts like [A_artist|B_artist|C_artist].

Also, I found an additional problem. After creating with DTG, if proceed with hires through the [✨] button, DTG is run again to create a new seed and a new prompt to apply. This causes little distortion at a low denoise strength of about 0.1, so it’s not noticeable, but as the value increases, the picture is greatly distorted.

and the "problem" is not problem
just don't use -1 seed.

[KohakuBlueleaf]

So I want to confirm if you want to:

• Only keep first part for DTG (I will make A|B back after DTG but DTG will not receive B here)
• Put both part into DTG (may cause problem if they are conflict)
• Get different prompt for the variations (impossible from what I can tell)

[namemechan]

가능하면 두번째 솔루션을 희망하지만 이미 충분히 노력해주셨고, 이 문법이 큰 이득이 되는것은 아니게 굳이 수정하지않아도 될것같습니다.

[KohakuBlueleaf]

@namemechan Now we have "no format" option which can avoid the problem related to these syntax(will not destroy them) but with degraded performance

You may want to check it

[namemechan]

Thank you for taking care of me while you're busy.