More wildcards

[MoonRide303]

I would like to discuss enhancing list of build-in wildcards here.

Requirements:

• it should be useful in context of Fooocus (proposed concepts should be known by CLIP model, so the conditioning could work),
• it has to be high quality (reliable source and good coverage of given subject / concept),
• it should be explainable and reproducible (another person using same logic could end up with same list),
• it should stay SFW (so people could use Fooocus in schoold or at work without getting into troubles).

As an example I will briefly describe existing wildcards that are already available in current version of Fooocus (v2.1.709):

1. color.txt - names of commonly used web colors, source: CSS 2.2 specification.
2. artist.txt - list of artist names recognized by CLIP model, source: Image Synthesis Style Studies Database (The List) (filtered out by CLIP).

If you would like to have more wildcards, please throw it into this discussion - just make sure requirements are met, and provide description similar to those 2 examples above.

[mmmz28]

Reproducible is important! When I use "color cat in a color hat with a color eyes,by artist" to get a style image that I like, I don't know which artist it is. The log.html has no record and is still "by artist", not the specific artist

[MoonRide303]

@mmmz28 I've added information about wildcards usage in v2.1.710 (#742) - if wildcards were used it will be printed in console log, like that:

[Fooocus Wildcards] input: __color__ cat in a __color__ hat with a __color__ eyes, by __artist__ and __artist__
[Fooocus Wildcards] output: red cat in a fuchsia hat with a lime eyes, by Adrian Tomine and Rinko Kawauchi

[MoonRide303]

And also in the Private Log, added in v2.1.711 (#743):

[MoonRide303]

And simplified a bit in v2.1.712 (most users won't be using wildcards and double prompt information would be confusing - so I removed raw prompt from the log).

[rubenet12]

Hi @MoonRide303
would it be possible to get the prompt used in the log when using wildcards?
I am using wildcards to mix animals and loved some of the results... but I can´t try to reproduce or work on that image without knowing which value was used.
I checked the log file as a text file and can´t find a the prompt version showing the actual values, just in case it was hidden.
Is there any chance to get a workaround or to include both raw and prompt in the log?
Thanks,
Ruben

[agarzon]

Just added a PR #748

I think it fits the requirements.

[MoonRide303]

@agarzon While ChatGPT is not really reliable source of information (it sometimes lies and/or don't provide complete information) it looked good enough to be merged - we can update it in future, if needed.