Link to Lora https://civitai.com/models/114798/blinky-pac-man
Does anyone have experience training a Lora for a really obscure character? Lol
I’ve done this a few times but creating this Lora was interesting so I wanted to write an article about the experience.
I got the request “Blinky from the Ghost Dance Meme” and I reacted Challenge Accepted.
Only 8 images and some images are low quality. This character is the OC design of Blinky from Pac-man created so there is really not much art to use.
With the magic of AI this is enough right? (Spoiler: the answer is yes)
Training Data
Blinky has these blue eye hair accessories that the AI really didn’t appreciate so I decided to tell the AI that these are simply blue goggles
and it kinda works!
She has a wide smug smile in almost all the training data so all my Lora’s output ended up with the same smugness so I consider that a win!
Note: I'm omitting the process of manually tweaking this images until I got decent results.
In the world of Art before AI, artists would create their own renditions of things that were not 100% the same as the original but rather an artist's personal interpretation of the original.
In that same way, my Lora isn't 100% perfectly the original but rather an artistic interpretation of the original. There are a lot of AI haters because AI isnt "Art" but if you are making creative decisions then how is it different from that so called "real art" ?
Too much philosophy?
Zero philosophy version: Its a woman with big boobs in a short dress wearing goggles that is based on pixel art of a red blob from 1980 = this is the power of AI.