A very much WIP LoRa for TikToker/Onlyfans performer Jameliz/justjellybeanss. This is the first time I've created a LoRa (or anything besides tinkering around with images in A1111 and i'll be the first to admit it's very weak. Now breaks much less often in v2 but can be difficult to get it to do exactly what you're looking for without other LoRas or ControlNet. Also has issues with lighting, even when prompting for night or dark scenarios. Will continue trying to fix and upload better versions.
Description
Immensely improved how accurate and how often it will produce the correct face. A bit overfitted, you'll need fairly strong prompting to get her out of backyards and houses and in clothing/styles you want. Recommend turning weight down to 0.7 - 0.8 when using other LoRas or if it's being too rigid. Restore faces + hi res fix used in example shots.
FAQ
Comments (2)
Have been going through guide to try to improve the LoRa. Running into two big problems, there's not much in the way of high quality stills of her I can find (most is from videos and introduces blur and noise into my newer epochs), and I am not sure how to best use Derrian's script to train them. If anybody has resources for these I'd really appreciate it :)
I haven't tried taking stills from a video, but doing some img2img hi-res fixes might help?