StylePile - a prompt generation helper extension with presets, variables, sequential/random selections and more

[some9000]

Check it out here: https://github.com/some9000/StylePile

StylePile is an extension for mixing and matching elements to prompts that affect the style of the result. Hence the name. By default, these elements are placed in a specific order and given strength values. Which means the result sort-of evolves. I have generated thousands of images (with regular SD 1.5) for each main Image type and tweaked the keywords to attempt giving expected results most of the time. Here are some examples on what you can get with a couple of clicks:

I left the less successful results in on purpose, to show that this isn't some magical tool that overcomes all the difficulties that Stable Diffusion currently has. But it does help you go on that artistic journey way easier.

Here's a look at what it adds to the regular interface (bottom left):

Ideas, feedback and help with the code are very welcome.

[some9000]

If someone with more coding skills/understanding than me would like to help this project out that would be great. Please check out this discussion: https://github.com/some9000/StylePile/discussions/8

Also, it's amazing to be featured in scripts section. But the name doesn't have a space, in case someone who can change that sees this.

[some9000]

This has received a massive update based on my own ideas and user feedback. Seems lightyears ahead of the initial version and, hopefully, will find a slightly larger userbase. Enjoy!

[grexzen]

This is awesome, can you please elaborate how is it different from the styles dropdown for positive and negative prompts?

[some9000]

Well, you get to assemble stuff right there from pieces, so to say. And some of the pieces have built-in negatives as well. Generally it's just meant to be quick and creative. Click, click, click, get result, change the dropdown, get a strongly different result. So it's "alive" unlike those dropdowns, where you just saved something static (unless I have no idea how styles work). The unfinished "Focus on" area is basically what the dropdowns you mentioned are. Just fused into one and, currently, made by me.

It would probably generate batch results for your SD-Chad easily to find the right combinations of keywords. Especially as I am about to add random selections to all dropdowns. Then, basically, just throw things at the wall and see what sticks.

In general I would just recommend trying it out. It's quick and easy.

[grexzen]

Oh nice, I created a huge list of those styles, but always have to either add more or go edit them before launching to change them.

[some9000]

If there is some element you often change that feels like it could be a dropdown here feel free to share. I picked the ones that seemed to have some general effect on the result. Like type, artistic movement, feeling etc. Perhaps there is something else to add.

It would probably be interesting to mix your tool for finding high quality images with something else to get their main keywords. Filter those and, basically, find the best keywords for each category. So far I did that but hand with thousands of test images.

[some9000]

I'm too tired to update the manual, but Beta branch has some really interesting features now that should be easy to figure out:
https://github.com/some9000/StylePile/tree/Beta

Almost all dropdowns have random selection now + there are even more hand picked artists + reasonably complex keyword mixing/matching. Still open to feedback/help.

[some9000]

I tried to convert this into an extension and, basically, ran into a wall. In case someone with more skill would like to do it, that would be super appreciated. It sounds like a better fit for the way it works, with examples etc.

So, if you'd like to help just grab the beta above and go for it.

[some9000]

Alright, good news, this has had a decent update. Finally works as an extension - you can just do this to check it out:

• Copy the main address (https://github.com/some9000/StylePile) of this repository,
• In SDUI go to Extensions tab,
• Select Install from URL and paste that URL into the URL for extension's git repository field,
• Press Install.

If a previous version was present as a script do remove it.

As far as I understand having proper tags shown for each image is not possible due to some issues with the current version of the UI itself: #4573 (comment) Hopefully that can be fixed some day.

[gvastethecreator]

Just a little feedback, i was using your script from the first versions ( which still have a version installed because it feels simple to use and helpful in some cases ).

Since the script basically adds different tag/keywords would be nice if we have a preview of which keywords adds to the original prompt.

Thanks for the hard work!

[some9000]

Nice to hear that it's useful.

As for showing the combined prompt, I had thought of something like that initially, but my Python coding skills were barely there. Now it actually might be possible to do. Will need to do some testing, see how it could work and if it feels ok.

Also, that doesn't really work with the batch/random parts of the script. So it may be a button that just fills the native prompt field with a full prompt or such.

[Inosss]

Hello @some9000!

I tryed your extension, it's very good, but i have one issue with it:

when i install style pie and select it from script dropdown menu, all images that i generate have no "seed" in description. I expect that with seed "-1" I will get the seed actually used to generate the image. But it is still listed as "-1". In that case i cant generate same picture twice, or i have to manually specify random seed in generate options.

[some9000]

Yeah, that is the thing I mention above about there being some problems with this. Even without the extension you wouldn't see the seed for every image when generating several. So, for now, this shows main keyword data.

Hopefully a fix will be available soon.

[Inosss]

Even without the extension you wouldn't see the seed for every image when generating several

@some9000 nope, i see all seeds if i pick one image from all generated.

[some9000]

Well, I certainly am not... Time to throw out all the other scripts/extensions and see if that changes something... Nope. And trying to use the function still produces an error here. Are you on the main branch of SD itself?

[Inosss]

Well, I certainly am not... Time to throw out all the other scripts/extensions and see if that changes something... Nope. And trying to use the function still produces an error here. Are you on the main branch of SD itself?

Yep. Naked SD with some extensions. Without them, it works same.

[some9000]

Alright, this has finally been fixed:
#4828 (reply in thread)

As mentioned there, I still need to modify an element of main 1111 code for it to actually show the different data for different images. But, I've way out another, that data is now there correctly.

[some9000]

Added some more functions:

• Found out that adding a broad "Concept" at the beginning of a prompt can have some fairly interesting results so that's now available if you choose an image type.
• A simplistic prompt generator is now available at the bottom for the times when you just want to test some looks, but cannot come up with an idea - feel free to suggest more prompt terms. You can edit these text files in StylePile\Inspiration subfolder. Clicking this will replace the prompt. At the moment these are super clumsy.
• Presets have been removed while I rework them into something greater.
• Tools and info panel that teaches advanced techniques by adding things into prompt.
• I felt the seed strength options felt clumsy so you can throw in [XS],[S],[M],[L],[XL] into your prompt and a certain string of characters will be randomly added to the prompt. From my research the more complex this string, the more variety will happen to images with same seed. So you can lock the random seed, throw, for example [S] into your prompt, set Batch size to >1 and see mild changes to the result. For this to work as expected you need to have this extra line to begin with. So you can just throw [XS] at the end of your prompt every time. To do this uncheck "Increase seed with batch size".

Here is the latest look of the extension: