Manga Colorizer ANIMA Edit Lora. Couldn't manage to add dual conditioning, advice if someone has made it work.
Hi guys, I trained this manga colorizer lora for the editing capability of ANIMA: https://civitai.red/models/2685547/manga-doujinshi-colorizer-anima-edit-lora?modelVersionId=3015470
It trained well, my first tests started getting good at about 2k steps. Then I asked CODEX to add a dual conditioning implementation for a secondary input image as reference for that lora. It wrote like 3 different methods and I ran a couple experiments, but in all of them the second image had close to zero effect, even replacing it with a white image didnt changed outputs, which confirmed to me it was being mostly omitted.
If someone has managed to implement in training and inference dual conditioning images (or more than dual) I would appreciate some advice on the method.
I have tried quite different implementation, Easycontrol in here, not yet supporting dual condition but implementing dual condition is quite obvious, concat additional input as cond.
have you tried https://github.com/wochenlong/lora-scripts-next/tree/anima-edit ?
Yes I ended up using that method for multi reference, I was looking what sorryhyun shared that they use a different method and the author says is more efficient, will have to test that one also.
I did a run with my dataset for two control inputs, and can confirm it is using both correctly, but I dont like the results because my reference dataset isnt good enough, might release it as a preview version.
