It's .. <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=7m9-UCuOfSE">The Ambiguously Gay Duo</a> (What's everyone looking at?!)Krea 2 v1: I updated the previous Pony tag-based dataset to captions so it can be used more effectively with the current models like Krea 2. A little sprucing up the image quality as well, but I did forget one important rule when it comes to captioning. Captioning the variance, not the qualities that should persist. So, unfortunately you'll need to be very prescriptive in your prompting to get the characters right instead of a trigger word helping as much as if I didn't dolt-head my captioning. Curses!Butt other than that, it' still performs well so I figured a slight inconvenience (especially since you can have VLMs/LLMs create a prompt for you and have it associate dark hair to Ace and brown hair to Gary, etc. - that's what I did when I generated most of the image prompts in the gallery). So if you want to know what a wrong captioning strategy performs like when you just have the captioning engine describe everything in the image, here's your example. It'll work.. just be prepared to be prompt-y.Lastly, no real need for bypass nodes/loras, or NSFW/penis LoRAs to get goonish level outputs. I've taken care of that for you within the dataset.trigger words that will help some: tagd-ace, tagd-gary, duocarPony version:Trained on some cleaned up versions of available images of the .. are they or aren't they .. duo plus a few more to round out flexibility. It shouldn't take much prompting to figure out which way they swing lolmain trigger words: tagd, tagdgary, tagdace, tagdcarhelpers: cartnon, 1980s cartoon, superhero costume, etc.

Files with SHA256: f4c5b99026c70038f915b6cbabe3e6f62f5aa23f52666b7f7ff5c2259cabe14c