Megumi ai model
Description
FAQ
Comments (4)
Rawr! Needs more natural curvy young women. Thanks for sharing!
Not sure if it's just the makeup, but it seems like this model's eyes are very very similar to your new Ayami model (and somewhat similar to Yolanda).
Thanks! Yes, I mix certain models to create some resemblances, though nothing too extreme. I might take facial features from one, the figure from another, and so on; that’s how I end up with so many models. This way, there’s a model to suit every user's taste.
BTW one thing as feedback. And that's for rendering without Turbo or six different plug-in LORAs.
If the required subject description prompt includes very specific things like the eyes, iris color, nose and such, it draws a lot of attention to these headparts in the render process. Which can sort of make it impossible to get exact camera angles. Or it increases the head size, possibly making the character look shorter or even dwarf-like.
Example in the one pic i uploaded where she's lying on her side in a mountain meadow, the head was notably bigger from prompting the eyes which looks odd. I basically uploaded the pic where the head was the smallest ;)
So normally i just ignore these descriptions to work with just the trigger word, maybe the hair- or skin color, though this Megumi character in particular was brunette and having a relatively normal figure without specifying that.
Point is that a character should already be fully trained with these properties. IIRC you should not describe these during training (like "Megumi with red hair and green eyes") but only describe stuff which isn't default (like "Megumi. She sits on a chair, and wears a necklace and a red dress...").
Otherwise, this character also renders super-sharp in a close-up. Worlds apart in quality from most models i've tested so far.
@LurkingAI @LurkingAI That is indeed how I trained it, but using five third-party LoRAs during generation always causes the hair and other fine details to break down; however, without those LoRAs, the photos end up looking cartoonish—which is why I use such long descriptions, even though a user could technically just use a single trigger word.