Type | |
Stats | 125 155 |
Reviews | (29) |
Published | Dec 27, 2024 |
Base Model | |
Training | Steps: 460 Epochs: 10 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | gaia_xiv, blue eyes, lips, black hair, black dress, boots, choker, belt, black thighhighs, bare shoulders, off shoulder, detached sleeves, platform footwear, |
Hash | AutoV2 0D437FC5D5 |
Gaia from Final Fantasy XIV.
New illustrious trained version. With new Training parameters!!!! I am getting ever closer to eliminating the need for the Negative prompt and baked in Screenshot style!!!
For the most part if you're using a non-realistic model the negative prompt is no longer needed!! Unfortunately, it will still show up in more realistic checkpoints.
Stick to the more ANIME style checkpoints.
Since my data set is rather limited, her clothing is over baked so it will show up more often than not. Since my priority was trying to get rid of the screenshot style, I left it as is. For the next version I will work on expanding my dataset so that the clothing can be changed easier.
______________________________________________________________________________________________
PONY Version.
Images generated with HiRes Fix. I also used Pony PDXL Negative Embeddings to help with the overall cleanliness.
My First attempt at a character model. I think I came out pretty good. I originally wasn't sure I was going to publish, but it turned out better than I ever expected. This will probably be the last new model for a bit. Next, I would like to go back to my other models and start working on them to increase compatibility with other LORA now that I have a better understanding of what I need to do.
The outfit is a bit over baked so adding "off shoulder, detached sleeves, ascot, jacket" to the negative prompt will help keep those from showing up on separate outfits.
.8 Strength also helps when prompting for non standard outfit.
Adding "((cgi, 3d model,3d):1.5)" to the negative prompt will help control the screenshot look when using other LORA
New to version 2:
I expanded the dataset but I'm not sure there is a noticeable improvement in quality.
The big thing is v2 should now be more compatible with other LORA
Enjoy and Good Luck!
-Daraco