This model was trained in 172 images using 172 custom image prompts. Originally there was a ton of issues since, my keyboard doesn't keep up with how fast I type. So, I used to spellcheck to help fix any grammar mistakes. I did include the dataset for download.
I would appreciate feedback so; I can improve the next build of this model. If you want to donate or whatever. That would be greatly appreciated. Since, I would love to get a better keyboard, so I don't have to spend a long-time rechecking grammar. -.-
If you do want to use paypal I do accept paypal gift cards if the gift card code is valid. I'm trying to save up for a $1300 computer so I can really get crazy with training a higher quality model. So, for now better models will very since, I have to stay within the 50mb training limit size. So, please understand for now. I have to train models under a pretty tight limitation. Since, this model had to be held back greatly.
Description
FAQ
Comments (3)
But what this lora is even for? Looks like for some bara artist based on your examples
Thats honestly a great quest. First, a chkpt model is basically the base model. It cannot distinguish the difference or reference characters. It's manly a model to be used as a backbone for ai art so data from machine learning could associate things like grass, water, fur, and so on.
A Lora model can be trained based on characters, objects, and style which chkpt models really can't. Using the pony model for lora models. You can get great results which is also very good for nsfw art while, allowing better training quality over sd1.5, sd2.0, and sdxl 1.0. When you start making a dataset, I strongly recommend you make as many images .txt file as detailed as possible.
Example 01.png will have a 01.txt file. If you can create a string of prompts like solo, male, standing (etc) with 20-30 prompts. The Lora model will be able to do better renders of said character. I personally try to do 30-50 prompts for very heavy party focused images so I can reference as much as I can. If you do a single character focused Lora model. I strongly recommend you use consistent art style of that character. An limit the images to 4 images for a front, rear, left side, and right side view.
Bara art can be done like you said with Lora models. Which I tend to favor when rendering art. However, if you specify the body type like slim, chubby, or something else. With a lora model the model will try to make the character look like that kind of body build. Thanks to how Lora models look to the chkpt files for non-character reference data information.
Sorry if that was kinda lengthy. But, there is alot different between chkpt and lora models. Even, with sd 1.5, 2.0, sdxl 1.0, and sdxl pony lora models. There is a difference between how and what they can do. So, it really comes down to how the model was trained for and for what purpose. Because I use furry chkpt models here they tend to favor more muscular art because, e621 and rule 34. If I used human male chkpt most of them are for muscular guys. So, with my Lora models you need to specify what body type you want. Otherwise, you'll get muscular art.
I did plan on doing non bara art but, I'm taking a break from ai stuff so I can catch up on some of my switch games XD
@FoxMccloud2022 But it loaded as LORA for pony, it doesnt look like a model, and the size is way too low for it to be a checkpoint.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.