New Lora Model!
Added 2023-09-07 15:45:56 +0000 UTCYou've probably heard me mention this a few times, but for people who don't understand this whole AI image stuff,
You have a checkpoint/base model. This is the model which is trained on generating the images as a whole.
A Lora model, unlike a checkpoint, you can't use it to generate images, however they are used to add more precision to the generation.
An example is the Kathrin Vaughan Lora I made a while back, my model was completely unaware of this character, and wouldn't generate it. With the trained Lora model, it could generate her with my model!
What does this all mean for you?
Well I've almost completed my data set for the TF lora model I've been working on, but before I go and train it, I would like to ask you lot, what type of themes or TFs would you like me to train this on?
Currently my data set has
TFs (Single and Sequence)
TGs (Single and Sequence)
(this goes for both TF/TG too)
Better M2F/F2M Genital Tfs
Magical themes with forced Tfs
better Paw Tfs (Ripped socks with paws coming out) I know people enjoy this one!
Overall Tfs will look more like Tfs lol
But if theres anything else you'd like me to add to the data set, please please please say!
Once a Lora Model is trained you cannot add to it, it must be retrained again
I'll be starting the training in a day or so, just to give people time to share there ideas
TLDR: What do you want me to train a lora model on to generate better images within a certain theme/animal?
Comments
While the Lora is being more trained on the style, I could work on one to generate better expressions, or research on some new techniques. I do love the idea of a costume TF and I will definitely explore the concept!
Solavine
2023-09-21 06:48:46 +0000 UTCI’d appreciate versions of your images where the character looks aroused in addition to scared or surprised. Or simply turned on, with no other emotions. Also, I‘m a sucker for the „Halloween costume triggers transformation“ trope. So that might be cool - maybe the costume involves a catsuit, a clip-on tail and painted-on whiskers? And then it all starts getting all too real…
Jakky
2023-09-17 21:23:55 +0000 UTCYeah, it's most likely because the original TF checkpoint I created was combined with so many models, a little bit of understanding had gotten lost. As I said to Zeta, it's quite bad for standing prompts and look more like post tfs, but seems to work best with prompts for the character to be on their hands and knees. Unlike how I trained my checkpoint, this lora model I will be going through and writing prompts for each image for the AI to train on. Currently I have about 150 images in my data set To give that some perspective the kathrin lora used about only 20 images.
Solavine
2023-09-07 18:00:46 +0000 UTCBetter transformation of the head! A lot of the time, the head is either fully animal or fully human, it would be really nice to see a blend of both! It's my favorite part of tf's, but I'm sure there's more people out there who'd also like to see that :)
Light
2023-09-07 17:06:11 +0000 UTC
