I'm not very familiar with Cyberpunk Edgerunners but I started watching the animated series and kinda just fell into a routine before I realized I started to like it.
This will produce the characters David, Rebecca and Lucy. Kiwi is in here almost, she needs a lot of attention but I didn't have many images to work with.
I tried for Dorio and Maine but they won't come through, those characters only produce a mix of the first 3 above it seems.
Some important details aren't coming through often enough, one of which is Rebecca's thigh tattoo, which I can fix later if there's interest in this but I haven't seen any Edgerunner LoRA so it doesn't appear to be of much value.
Anyway, I like some of the characters so I made a LoRA.
With this single LoRA you can generate multiple characters at the same time, and have them interact.
This is a SFW version but I generated quite a few NSFW images that I left out of the training, though you might be able to distort it otherwise, I haven't tried so I don't know. If there's any interest in an NSFW version then it shouldn't be too difficult to augment what Krea2 can already do.
Description
FAQ
Comments (7)
Finally, I'm glad to see Cyberpunk content on this site. I hope you continue to improve it. Also, one of the things I've had trouble finding here is Wuthering Waves characters. I don't know if you're interested, but there are so many characters. If you consider it, I would love to see it included.
Sure, I've got nothing but time, I'll have a look.
@Enigma38
So I looked into it and found 57 characters associated with Wuthering Waves. I actually wouldn't mind doing all of them, here and there during down time, but gathering the images (dataset) is a chore and not something I'd slice out time for so, if you have a few characters that you particularly like then I could probably get those done, especially if you point me to a nice, packaged source of images for those, 50 each is not a bad number, containing profiles, closeups back/front views, full body and various poses and expressions. If you provide the images I'll definitely get you those LoRA.
@Tarterbox Thanks, I'll work on it. You said 50 images for each character, right? I hope I understood correctly.
@Enigma38
Yes, I said 50. per character, but I don't mean it, it's just a quick way to get to the data. Here's why...
There's unlikely to be 50 usable images of any character, so if I ask for 50 I get everything that's potentially usable, and from that I can pick and choose which to use for training.
I've used 4, 6, 8, 12 and 16, as well as hundreds of generated images for training data, depending on how versatile I need the model to be. A good dozen images, with proper angles, and many of the extreme poses and expressions that one would want, is plenty.
Models generally are able to interpolate between angles so we don't need EVERYTHING under the sun. Basically, if you want it in there, include it.
I love doing this stuff so it's no bother, I'm learning quite a bit about this particular model as a result and if someone else is doing part of the work then I'm a happy frog.
@Tarterbox Now I understand 😊 I have one more question. If I convert these characters to real-life characters using the Anime2Real method, would it be possible to combine both their 2D and real-life appearances into a single LoRa file? If so, I'm willing to work on that as well.
@Enigma38
I made a LoRA that was able to produce a dozen or so pretty esoteric styles, each one visually discreet, but none of them verged on anything close to "real" looking, and all had a very synthetic, fabricated feel. So I can't answer the question with any confidence but I don't mind giving it a try.
However, what you're talking about doesn't seem necessary since it's a modification of style where using another LoRA, or just prompting, would be able to accomplish one or the other.
I would test first, one style, then prompt the other or use a modification LoRA, to see how it fares. From there we can try other things. One thing I worry about is that an attempt to overload the LoRA will reduce versatility (prompt adherence). But if you feel up to it there's no obvious reason not to be prepared and gather the necessary data anyway. We can always try both and see which one works best.