Description
In the process of being perfected
This is my OC, Glitch Flare, an Alicorn with a ‘fiery’ temper who loves video games. Not surprisingly, her Cutie Mark is a gamepad. And this is exactly my main challenge: I can’t seem to train a LoRA that correctly recognizes the custom Cutie Mark I made for her. I’ve tried everything:
- I manually applied it to all the dataset images using Photoshop.
- I cropped portions of her flanks and trained them in a dedicated folder (Detail Cropping).
- I even tried training the LoRA using the standalone Cutie Mark logo image…
Nothing works, the AI always distorts it or generates a mess.
So, if you guys have any advice or tips on how to proceed, I would really appreciate it!
Currently, I’m training a ‘clean’ (Blank Flank) version of her LoRA, which will be available soon.
Currently, I’m training a ‘clean’ (Blank Flank) version of her LoRA, which will be available soon.
Source
not provided yet
I understand what you mean but the problem is not the tool, but who uses it, I use Local AI (also Textual AI agents) because I hate limitations and censorship.
Generating
I’ve used perchance ai, since it is free and has no sign up, it’s quality can be surprisingly good at times, it’s users aren’t the greatest tho, some decent and good people on there but got real bad pedophile and degenerateproblem.
Are you talking about training or generating?
Why do some ai generaters have much clearer images and less pixel depth?
Foggy
Thank you very much :) Now I’m working on the Cutie Mark which the AI really doesn’t want to learn, or rather, it has understood where it should be placed, but it always distorts it.
I also wrote an article on Civitai: https://civitai.red/posts/28870517
Edited