This LoRA can be used standalone, but it was specifically trained on 1000 unique faces to un-Krea faces and produce more variation in facial expressions across an image with more than one subject. You can prompt all kinds of face details and facial expressions and they will obey the prompt. The sample images were generated with my Amateur Lesbians LoRA and edited with a simple InPaint workflow.
Description
This LoRA can be used standalone, but it was specifically trained on 1000 unique faces to un-Krea faces and produce more variation in facial expressions across an image with more than one subject. You can prompt all kinds of face details and facial expression and it will obey the prompt. The sample images were generated with my Amateur Lesbians LoRA and edited with a simple InPaint workflow.
FAQ
Comments (5)
Can't wait to try this one out. The examples look great.
experimenting with this and im actually getting more facial variation than i am from any other thing i've tried. it even helps more than seed variance enhancer nodes - although i've only been randomly messing around with those (no idea what i'm doing).
i am SO DAMN SICK of the standard krea faces everywhere, especially that godforsaken default blonde female face. a perfect model would be the prompt adherence and knowledge of krea2 + the random face and body variation of qwen2.1
I was at pains to make sure I included as many different facial expressions as possible. I'm glad it's working out for you as well as it works out for me. Show me what you've got.
@BBW_4TW it didn't pan out as well as i thought initially. your lora still helped, but im really starting to doubt it's possible to fix krea2 in this regard; i just keep noticing the same things over and over everywhere. not to take anything away from yout lora ofc.
@BBW_4TW like seriously... my workflow is: Krea2 RAW (avoiding fine tunes) -> SigmaSync SDA LoRA on the first two steps -> multi GB refusal reduction lora -> raw to turbo lora -> random lora combination in between with hopes of achieving smth -> seed variance enhancer -> get the same face + body anyway lmao.