Stable-Diffusion: Advanced Captioning (Booru Dataset Tag Manager)

[@TheRoyalSkies]

Here's the link to the BooruDatasetManager: github.com/starik222/booruDatasetTagManager And, if you have any other questions feel free to drop by our Discord under "AI-Questions" here: discord.com/invite/TdkyE9yZZ4 Hope that helps, and see you in the next video-!!

[@verse369]

Things are getting complex, but it's going to be worth it!

[@lefourbe5596]

For realistic character/AI model you should use the ancestral CLIP caption style (ahra, there is a woman standing in front of a green screen, wearing a Fantasy outfit...). What WE use is TAG style (ahra,1girl,blonde hair,...) that is also compatible but it is best used for anime model usually if they are trained from danbooru. (Hence why the captionning tool is called like that). Some says to uses a mix of the two. Now they might be better tool to manage full sentence... But i'm not familiar with them and i do it manually with the bare minimum description if i have to (as we see on the civitAi exemples) Idealy, I advise to do less that 10 tags : Name, class, body position, outfit, face expression, background, style.

[@kernsanders3973]

He will probably be using a base anime model to train which should have booru tags trained into it's textencoder. He mentioned it in previous videos he is making an anime character. Either way I've found that booru tags work okay on none anime models for prompting, so theoretically training it like that should be okay.

[@lefourbe5596]

@@kernsanders3973 well i know cause i basically taught him everything :). target will be AnyLoraAnimeMix model (beter to train base Anylora but it's for simplification)

[@AnnisNaeemOfficial]

@@lefourbe5596 thanks for all the info. just to confirm, you're saying don't even try doing dreamshaper or anything, just go straight for the anylora model?

[@lefourbe5596]

@@AnnisNaeemOfficial well both are okay to train but i found that Anylora training work on most anime model. that is a rare thing on realistic model and SDXL ones

[@AnnisNaeemOfficial]

@@lefourbe5596 I’m training a cartoonish style. Not anime. Western comics. Still advisable to go with anylora? Thank you

[@Gounesh]

Hey mate! Thanks, as always! Hope you can make one for style as well.

[@Lv7-L30N]

very useful, thank you.

[@darrennew8211]

This is really cool. You could use this for things other than characters, right? If you wanted space ships or castles or something instead?

[@TheRoyalSkies]

Oh yeah, absolutely, we're using a character in this example, but you could just as easily use this for environments and props as well!

[@ZhiwanCheung]

Hi. Great video. For the prefix captions, ie (ahra,woman, green background), should you then have these three in the prompt as well when generating your image? Or is the keyword "ahra" enough? How should we decide what goes in the prefix?

[@caseyj789456]

1:19 noted! :)

[@amoawesomeart6074]

fort the life of me I cant get the Koya script to work, I can see it downloading the models but than it just refuses to run.

[@n.bcritical9819]

why the hell am i not getting notifications for these videos ?!?! i have to manually check for new uploads

[@mat5844]

Has anyone managed to install this on Linux or macos?

[@lucasouzah2a]

I don't understand anything about these videos you're posting! I don't know what the purpose is and where you really want to go with them! I'm more lost than blind in a gunfight!

[@lefourbe5596]

to train stable diffusion on your picture ; install kohya ss make a description of each picture with it (utilities tab) fix the issues manually with boorudatast tag manager save ready to train for the next video (this will be ven harder :p) could be used for automatic coloring tool for 2D drawing, stylization of your subject to be coherent in a project, changing clothes easily, or just quick concept around the subject. i like to do emojii with it.

[@TheRoyalSkies]

Sure, this video is part of the AI training series - It's to help artists create their own datasets and train AI on their own art, but it wont make a lot of sense if you haven't seen the ones before it lol - Here's the complete series if you're interested, but if you're not no worries, we'll be back in Unreal stuff after this series is over :) kzbin.info/aero/PLZpDYt0cyiutYOWX9_gHR6YyT9Y0JHByF

[@lucasouzah2a]

@@TheRoyalSkies I watched the videos, what I didn't understand is where it will get you, what results you will have after following the steps and if it's really worth it! I don't know if I'm going to do this, but I'd be happy to see where I'm going, to understand what I'm seeing! Unless of course the videos are some kind of game and the end result is not really what I expect and showing this now would make us stop watching the series of videos!

[@Lv7-L30N]

@@lucasouzah2a maybe if you dont need references to make your projects... you could skip this