[Feature Request]: Improve docs for styles

[ppbrown]

Checklist

• The issue has not been resolved by following the troubleshooting guide
• The issue exists on a clean installation of Fooocus
• The issue exists in the current version of Fooocus
• The issue has not been reported before recently
• The issue has been reported before but has not been fixed yet

What happened?

There are SOOO many style buttons, but not nearly enough documentation on them

I'm calling this a bug, because the current state makes them almost unusable.

The tooltip preview images are kinda nice, but:

• Some of them arent really accurate. Half of the ones with "anime" in their names, dont have an anime-looking icon. So they dont match what is actually genereated

• There are multiple groupings, such as "SAI", "MRE", "MK". but zero explanation of what they mean

Steps to reproduce the problem

open the style menu

What should have happened?

more details.
Maybe a toggle for "show buttons as icons", so you dont have to Individually.Mouse.Over.Each.Button

What browsers do you use to access Fooocus?

Google Chrome

Where are you running Fooocus?

None

What operating system are you using?

Windows 11

Console logs

N/A

Additional information

No response

[mashb1t]

@ppbrown renamed and retagged as feature request, as nothing concerning this request is broken.
Previewi mages have been created by a standardized method, namely seed 0 with only "cat" as a positive prompt and the single style activated.

[ppbrown]

Just because there's a standard method of generating them, doesnt mean it's the right method!

if you start with the anime model used by "run_anime.bat". and then you click only the "MRE anime" style.
and then you prompt for "cat", with seed 0...
IT LOOKS NOTHING LIKE THE ICON STYLE!!
Not even vaguely . The current style button icon for it looks photorealistic!!!

Even using sdxl BASE MODEL with seed=0, prompt=cat, and "MRE anime" style selected, gives a more anime looking image than the icon for the style

That shows that the current icon generation method has a bug in it. For pretty much all "anime" style buttons.

[DavidDragonsage]

There is a very nice display of all the styles here:
https://docs.google.com/spreadsheets/d/10UXU59fBA5wZrcEEvgv1MTf1MSNTCB5PT4fNxsAst8E/edit#gid=2125777153
That kind detail could not be reproduced in a popup image

I see can see your point, but the current system ensures that the style itself - without the influence of a different base model - is always shown. There are advantages and disadvantages to this standardization. I think there would need to be a community consensus to change the current system.

Most of the included styles would look quite different if used with the animaPencilXL_v100.safetensors model, and quite a few of them would actually be quite useful with anime. For example, the new Fooocus Semi Realistic style was targeted for anime and other non-photorealistic genres.

The cat for MRE anime may not look particularly anime, but it is not photorealistic. The style is actually having an effect on the default base model.

[ppbrown]

Thanks for the link. thats really helpful.

i think there are multiple problems with the current display:

1. you have to mouse over each one. This makes it virtually impossible to visually compare multiple styles properly before rendering. When rendering can take 30 seconds, this is an unpleasant user experience

2. the photos are too small in-app

3. a cat may be cute, but it is a TERRIBLE style base for comparison. The "redhead on beach" in the page you showed, displays the contrast of styles much, much better than the cat, in the same grid.

You talk about getting a community consensus before making changes to this.
Okay, so how does one go about officially getting a consensus?

[DavidDragonsage]

Thank you for providing those additional details. I really like those cats - when I think Fooocus I think cats - and I don't even have or want a cat! But a person may well be more practical.

But then I see we could get into questions of ethnicity, and this is very much an international product. For example I would vote for Afro-American woman because then I can easily see the differences in hair detail - and therefore overall detail - that are influenced by the style. I use Afro-American (actually Afro-Canadian) for testing base models, LoRAs, styles and Fooocus settings such as Image Sharpness because of this.

Actually I can see that putting a person in the popup image could be a real can of worms because of ethnicity. That may be why they chose cats. I'm confess I would have to concur on this choice.

But the other question is whether the images should use different base models if the styles are intended for particular genres. Is that another can of worms? I do a lot of fantasy work, so it would be useful for me if the cats for the fantasy styles were based on the Faetastic model. But I've gotten pretty used to what the styles can do, and basically I use my own images as a reference if I want to create yet another magical portal!

As for forming a consensus, I don't really know. I assume its by seeing what kind of responses they get to a feature proposal like this one. I've noticed that some of the proposals go through quite a bit of discussion.

Thank you for opening a thoughtful discussion - hopefully some other folks will join in.

[ppbrown]

A person really is the best choice.
Maybe if allowing for diversity is a major criterial, then the best option is simply to offer people a choice of what type of style icon to display:

1. cat. 2. caucasian female 3. other gender/ethnicity. 4. ....

but since 95% of models are going to render "eurasian female" in everything, there is a question just how useful that would be to have a whole bunch of different PC-compliant pre-rendered icons.

Perhaps the ULTIMATE feature, is to simply give people an option to auto-render THEIR CHOICE of subject, to generate their own icon set for the buttons, locally on their machine. (ie: allow them to type in a prompt of whatever they like)

This becomes even more useful if there is also an option, "generate and save a larger-format grid". Sort of like that example matrix you gave me.. but just for the one representative subject across all the styles

[xhoxye]

The preview is only a reference, it does not fully show the characteristics of a style, it should not be based on other specific models, preferably based on the official SDXL 1.0.
With animaPencilXL, some styles will be invalidated.

Also, I see another kind of UI that is displayed in the plugin for webui
https://github.com/javsezlol1/Stylez

[ppbrown]

"How do I know what style button to use?" is a fundamental, basic question.
Making it usable, should not require an optional plugin

[drdancm]

Could we simply get a link inserted near the "Styles Heading" to the following location:
https://docs.google.com/spreadsheets/d/10UXU59fBA5wZrcEEvgv1MTf1MSNTCB5PT4fNxsAst8E/edit#gid=2125777153

No PopUP needed, just this link.

It would hardly take up any space, and hardly any work, No consensus needed because it would not slow anything down, and if you didn't want it, you would simply disregard it.

I wish I had heard about this before. Just shows how a thousand pictures are far more informative than many style names.
Fooocus is the pre-eminent Easy to Use for Beginners, but powerful enough for almost anyone else too, hence anything that helps clarify what various option actually do would be totally in keeping with the "Ease of Use" approach to Fooocus".

[DavidDragonsage]

drdancm - I agree, that's a great resource. I refer to it a lot and for me at least, its all I need. I think your built-in link solution is a great idea!