- Jun 9, 2019
- 432
- 1,612
Basically, if we want to you just switch faces and stuff and just make a toolkit for generic scenes. This way you can switch between them all for different things so we don't need a giant render machine to do it. We just hire a few artists get some good art.
- Mar 30, 2019
- 92
- 115
Absolutely.
You can define poses, lights, backgrounds and everything...
But then you don't get that specific KoD Artstyle. It would end up like one of Disneys "live action" remakes of animated classics -- which never work.
I tried to model KoD Characters in Daz with or without a "cartoonsh" style years ago, but that didn't work ...
You don't have permission to view the spoiler content.
Log in or register now.
With AI you can relatively easy achieve 80% of desired image quality, but it nearly impossible to beat other 20% (with AI tools, because they also not give full quality), always some small details are out of control. And AI is so bad at interiours)
You don't have permission to view the spoiler content.
Log in or register now.
Last edited:
You don't have permission to view the spoiler content.
Log in or register now.
I don't like DAZ, because this render is very constrained in styles. Blender provides more flexibility in styles, also you can combine 2D and 3D graphics.
- Mar 30, 2019
- 92
- 115
It actually would be interesting detail to build in a world tendency in KOD which changes the art style for the scenes and the dialogues opening up more possible and different scenes + writing. Then guides and exploring also would make more sense to see all different options
- Mar 30, 2019
- 92
- 115
It will be a steep learning curve, especially, when you don't have much free time to learn such tools. I haven't even finished "the donut".
Are there any good tutorials around character posing in Blender?
- Jun 19, 2018
- 21
- 24
Playing with illustrious checkpoint:
Dynamic poses is awesome)
You must be registered to see the links
Dynamic poses is awesome)
Last edited:
I tried to do OC character LoRA. It take half a day making consistent images (not so hard with reinforcing and doing a little tweaks from time to time). Main problem is that LoRA can bleed artstyle in model too, but its not like that a big of a problem for making a game honestly. Like, I'm a newbie in this, and start tinkering with AI on site when heard that game died. So, only like 2 weeks of expirience and still got decent result in consistency of generating character (tried in various styles and checkpoints based on Pony).
You must be registered to see the links
I would like to use FLUX because its ability for object-oriented syntax for better control of details and actions (as much as i tried ti use it with different checkpoints, only FLUX provided consistent results with it, and while still not even close to perfection, level of control on complex image is on another level), but ultra and pro versions seems to not take LoRA's on site, and my potato computer take forever for anything around stable diffusion. Real shame#2, as there is no inprinting on website version.
And while I would really like to see remake/continuation/spin-off, I'm sure it would backfire. Like, Nomo doesnt seem like such a good person to, returning online one day, just to accept that smebody else doing their own version of his project. It would be great shame to see better iteration die just because of spite of some lazy ass.
So, to take general idea and to do another version with new characters should be safer, and allow for more freedom in writing.
And like, its not that big of a problem to get orcs or minotaurs in AI generation. To create fully new race -- that would be definetely pain in the ass without artist, but probably still doable to some degree.
To get actual artist that can make his own vision from scratch and to polish AI-generated basis would be ideal, but by being not a lazy ass with AI, some good stuff can still be made.
FLUX barely supports NSFW content. At max you can generate boobs and thats all what you can.)
yeah I know, but at least with better syntax its easier to make there some specific concepts probably, but I'm not sure if it would be effective cost-wise. FLUX cost too much to use (even some mundane scenes was hard to get, and it was mainly easier to get a shot with 2 chracters without messing attributes between them), and it can be easier to brute-force with Pony or some NSFW specific checkpoints concept of character, rather then to get enough material from FLUX (there is no such a big problem that require syntax in solo art, even if it is a good instrument for specification of image). As I said, I'm newbie in this.
Also, can you please pinpoint me to how make models understand specifics regarding face proportions and lips form? I believe it doesnt seem to understand me when I try to use "tall face" or other wordings for little to none result. I'm really annoyed of receiving another bunch of kawai face, and instead of more mature and regal face, I'm bombarded with another bunch of round ormore infantile faces, it takes too many iterations to make models understand what i want. I tried to find some guides and libraries for key-words, but alas, didnt find anything useful for my problem. Models continue constantly jump between MILFS/GILFS and Lolis by face.
I have same problems with Kira and Riven and i don't know how to solve this problem.
All i can suggest to you is to use ControlNet for more coherent generations.
Eh, as I said, my PC is piece of potato, and any 1 try take 1-2 hours to generate 1 image. Like, its easier to me to just brute-force it and maybe make LoRA, I'm satisfied with first try. Same wuth inprinting -- tried to use it, but too time-consuming. If there is no online service with such instruments, my hands is tied here.
Trained 2 illustrious models for Riven:
A: trained on original data, from game;
B: trained on upscaled data (non-smooth bicubic interpolation);
B model is better for me.
UPD: it sucks with other styles(
A: trained on original data, from game;
B: trained on upscaled data (non-smooth bicubic interpolation);
You must be registered to see the links
You don't have permission to view the spoiler content.
Log in or register now.
B model is better for me.
UPD: it sucks with other styles(
Last edited: