Update: For an unknown reason, all my tests on-site kept failing, i have re-uploaded a pruned version of the LoRA where any unusual layers has been removed but it still doesn't work 😭 So, i have disabled on-site generation 😥<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d6f12357-e229-4421-9199-e3e1a5ca5962/original=true/d6f12357-e229-4421-9199-e3e1a5ca5962.jpeg" />When using generated content for training, you can sometimes get stuck in a cycle of finding what should come first: the egg or the chicken? Well, here is the dinosaur that laid the first egg for my UnNamed upgrades.Based on some pictures made by @daceheg192491 for <a target="_blank" rel="ugc" href="https://civarchive.com/models/2842144/krea2-dmergev3?modelVersionId=3214144">DMergeV3 </a>(highly recommended checkpoint!), i captionned some of his work (to avoid re-using prompts), generated a few pictures using Krea2 and some slight img2img using UnNamedIXL, did a training to get a first LoRA, merged my Ai Dolls for Krea2 LoRA in there together with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2727829/krea-2-enable-nsfw-prompt-adherence-krea2-nsfw?modelVersionId=3066310">some anti-refusal LoRA</a> and finished with this pretty broken and far to "wet" model very focused on glamorous women.This allowed me to generate the dataset used afterward for training, but in any case, here is the source LoRA for anyone curious to see it 😉💜

nKrea - Krea2 LoRA experiment

Posted on Civitai, upload found elsewhere are FORBIDDEN and done without the author permission.

Files with SHA256: 8af27f0f804046751c2380be0c69e37c5ae39660fb3397e8a4e50d46064ec78e