Tegan is trained from 100% generated images! Any likeness to person(s) living or dead is as coincidental as finding your own doppelganger out there - could happen, not intentional. Her reference images were created using Krea 2, Flux 2 Klein, and Gemini.
Just describe a scene with "Tegan" in it and she comes along with her red hair and freckles
Glam vs Girl Next Door
Glam is the first version. I was able to get a consistent body generating in one tool and a consistent face in another. I set up the images, captions, and training to build the model. I like her, but she is too "AI Woman" to me.
Girl Next Door is the version I started making as I uploaded Glam - having stared at hundreds of test runs and adjustments, it needed change. I wanted some asymmetry in the details and a little more down-to-earth/real look, so light blemishing, more detailed hairline/scalp, nostrils, eyes, and lips. Mainly, I regenerated and adjusted her facial structure and expressions to feel more human.
Both models use the core body style (each has proportions tweaked differently via LoRA merges) - so freckling and detailing should be similar.
Description
This version has a hand tweaked face and improved hair and emotions
FAQ
Comments (6)
Very, very impressive, this is easy one of the best Character Loras I've seen so far.
First off, she is extremely pretty, though that's subjective, of course. What is objective, however, is that this LoRA is exceptionally well-trained, with a perfectly curated and massaged dataset. The skin texture is outstanding, and the versatility (responsiveness to prompts for hair and eye color, facial expressions, etc.) is top-notch. Hats off, keep them coming, and above all, thank you so much
Thank you! Feel like I nearly burned out my local rig making data sets!
Edit: GND version is back online and she is generating WONDERFULLY! Do enjoy :)
OLD:
Because I hate to have more than 6 buzz at any time, apparently... Girl Next Door version will be re-uploaded tomorrow after a re-train. It was getting strange facial breaks and uniboob (likely do to the merged LoRA). I have rebuilt the training data to reflect, so the few of you with the GND version are likely to see a general change to her, but all the core markers and structure SHOULD be there (if my training data serves me well).
In for the article on character training! Great LoRA.