Trained on AnyLora
To get the best results, please read the description below:
Recommended weight: 0.6-0.8
Activation word: Nova
Upscale to 1.5
Using anime SD models is recommended, I haven't tested any semi-realistic checkpoints.
For eyepatch use: yellow eyes, eyepatch
For heterochromia use: heterochromia, yellow eye, blue eye
I would appreciate any reviews, positive or negative!
I would very much appreciate you sharing your results ^_^
If you have any suggestions on what LoRA i should do next, please share them in the comments
Enjoy :)
Description
Version 1 - trained on 65 images
FAQ
Comments (17)
Calypso plz
Next one's gonna be Alex, so perhaps later :) Will do though
@12user34kn276 ohh good to know. Alex is also cool I like her UwU
@12user34kn276 NICE! Will you do Once In a Lifetime as well?
@KingLewyin Haven’t played it yet so probably not :(
@12user34kn276 its made by the same dev it was his first game while the quality isnt the same as eternum, its such an amazing game id highly recommend, ive played hundreds of vn's and OIAL is easily top 5
@KingLewyin In that case i’ll definitely check it out!
Well, you got the eyes and eyepatch better than my training attempt did. Care to share some of your tips?
If you followed the tutorial, i believe there are only three things that could make the difference - the dataset itself, training resolution and number of steps. Remember that the eyepatch and eyes are asymmetrical elements so turn off flip_aug, and make sure you have enough steps. Finally, to improve the details you can try setting the training resolution to 1024.
@12user34kn276 Did you use "heterochromia, yellow eye, blue eye, eyepatch" as tags in your dataset?
@misterandy33 heterochromia and eyepatch were among the tags automatically assigned to images in the dataset
@12user34kn276 Which tutorial are you using? I can never get any sort of enhanced specificity automatically from the one I'm using...
@misterandy33 Sorry, I’ve mistaken you for someone else at first whom i have recommended this tutorial to make a LoRA of Nova: https://civitai.com/models/22530?modelVersionId=26903
This tutorial provides a collab page which I always use for my LoRA’s
@12user34kn276 No worries. That's actually the one I was using... But I actually just tried using DeepBooru on an img2img, instead of CLIP, and got more of the kind of tags I was looking for. I'll try using that for the dataset later. Thanks for your help so far!
@misterandy33 Cheers✌️
Best eternum girl
I'd say it's a tie with Luna. The Lora is gorgeous, tho