V2 is a photo/detail slider that was trained directly on SNOFS Krea 1.4. Krea 2 was trained on a ton of artwork, especially for sexual things, and so the model likes to pull in that direction. I've lessened this problem over time and will further 'fix' the issue but fro now, this works wonderfully. The range is about -3 to 3. Negative values will pull toward illustration/anime. Higher values will pull towards photos.
Description
FAQ
Comments (6)
Thank you! You have a super models, but anime effect was annoying
It's a bandaid for now until the model is trained further, but it works quite well.
@Ashen3 will wait for the next updates. you're awesome
thanks for this, how far the negative and positive can go ?
Can? What's the max int for the field? I'm sure it's at least in the 10,000's. haha. He said it kinda works like a slider but he doesn't know much about making sliders. Meaning you'll just need to experiment.
Highly recommending this LoRa! It's positive impact on photorealism is outstandingly good and you also get a great face variety when using this LoRa
I use it at the end of my LoRa Chain with 0.85 strength: Snof 1.1/1.2 with 0.55 strength, nicegirl LoRa with 0.55 strength and finally this LoRa with 0.85 strength.