This DoRA attempts to replicate the art style used in Xenogears. Note that it doesn't attempt to make any specific characters; however, it will probably reproduce specific characters when combined with character LoRAs. This is for Envy Starlight.
Description
FAQ
Comments (7)
At least someone remembers the wonderful game
this game is legend, better than all of final fantasy series. Its actually designed as "Final Fantasy 7" until it was scrapped to other name.
I wish they'd been given enough time to finish it properly.
Single question, could you make a tutorial of how to locally train DORAs? Not talking about settings, more about which things we have to install to run it.
I train DoRAs with kohya_ss, which you can find here:
https://github.com/bmaltais/kohya_ss
Then I use this code I threw together to zero out some of the layers of style DoRAs:
It may be possible to do this with SuperMerger in A1111 now; I haven't checked lately. That's where I lifted some of the code in there from. In any case, you can run it in the same environment that you run kohya_ss in, although you'll need to install node.js and a few node packages (it'll complain about the missing ones, so you'll know what to install).
@_Envy_ Thanks for the answer. I see the code.. do I need to replace it or something or just python launch it after configuring it?
Replace the input and output filenames of the LoRAs in the file, then just run it with Python. For the layer weights variable, I suggest one of the following:
0,0,0,0,0,1,1,1,1,1,1
0,0,0,0,0,1,1,1,0,0,0
0,0,0,0,0,1,1,0,0,0,0
The more layers you zero out, the cleaner the final result will be, but the less of the training it will carry over.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.