In celebration of its 30th(!) anniversary, I thought I'd do something fun to celebrate my favorite JRPG of all time! This model was trained on about 100 screenshots from the anime cutscenes made for the PS1 port (later used on the DS and other ports). It's not 1:1 with the cutscenes, but I think it looks good, and I'm pretty happy to represent one of my favorite games of all time! RIP Akira Toriyama, rest easy knowing your work on Chrono Trigger resulted in one of the greatest games of all time!
Note: some of the seven main characters have minimal screentime in the 10 minutes or so of cutscenes. For example, Crono, Frog, and Alya (best girl!) have a good chunk, while Marle, Lucca, and especially Robo and (SPOILERS for a 30-year-old game) Magus had slightly less. Still, you should be able to make images of all of them (Robo is questionable), even if outfit details aren't always 1:1!
Description
FAQ
Comments (9)
Some bleed of features and tends to shift some colors, but overall works pretty nice and those issues can likely be corrected with weights. I like it.
Yeah, this one was a bit tough to make, as there was limited footage to work with. Still, all things considered, I think it turned out pretty decent.
I'm curious since I don't think you made an article on it, but how do you make a lora like this?, i.e, a lora that is both a style lora and a multiple characters lora.
So, this followed similar principals to my multicharacter LoRA. For the characters, I just manually tagged them with their names (i.e., Ayla (character). Otherwise it's pretty much the same as you'd normally train a model. If you're making these directly on CivitAI like I usually do, you will, of course, want to use the Style setting. If using Kohya/training locally, you'll largely just want to follow best practices/look up the best settings.
@metalchromex I see. So no need to make a dedicated folder for style right? That's the part I was most confused on since you've shown that folders make everything better in case of characters.
@Silvererased this model was actually a bit more straightforward than most of the other models I made. I had just shy of 100 pictures. What I did for this was just run the auto tagger (TagGUI) on the folder to automatically caption all 90-something images. Then, I just manually added '[character name] (character)' to all instances where that character appeared. At that point, it was just the standard "zip up the single folder and train it."
By technicality, this is a DBZ style lora.
Kind of? I'd say it'd be closer to an Akira Toriyama style if you frame it that way, but yeah, DBZ by extension.
Came back to this after a while cause of another similar LORA and apparently TYRONE, Asthore, whoever the deleted user was, and Kyokatako are unaware that DBZ and this game's artist are the same.





