Fan Art Kingdom of Deception and Noxian Nights [fan art, ai models, waiting for updates, future plans, healthcare]

5.00 star(s) 1 Vote
Discussion Reviews (1)
K

KGARrokh

Member
Oct 18, 2018
145
461
Vyacheslav Grinko said:
This thread is even than the actual game I'm mind blown how good this images are
Click to expand...
I actually mostly dissapointed about image quality (in general), only 10% or less of images are worth something, other 90% is dissapointment (in my case). But sometimes you can get something really interesting)
ZV5KPJHTVHHH9CEN75M37ZTC30.jpeg W10FET02DQTS4HN983ZYXTWWW0.jpeg 7AWQ4H7KJ81YW9KPVC2SWM66G0.jpeg CMECZREVJBS1D7NBFQT63ATN60.jpeg A28V71WMWNPW6VBK4V6RPZTCZ0.jpeg VA4T59KFDZ1DC2P08EYZCKPET0.jpeg
You expecting highly detailed perfect image, but almost always get image with defects (bad arms, missing arms, deformed eyes, absurd background, bad coposition, perspective distortion and so on). Something always wrong)
 
Last edited:
  • Like
  • Thinking Face
Reactions: true_exorcist, Dante477, Eprofesy and 4 others
K

KGARrokh

Member
Oct 18, 2018
145
461
xyz_grid-0001-494309299.jpg
I've maided first attempt for single character model for PDXL6, SDXL1 and Flux 1. D. Current i've done tests only for last epoch for PDXL6, my rating for this model is 5/10. Link:
You must be registered to see the links


You don't have permission to view the spoiler content. Log in or register now.

Other epochs and models for SDXL, Flux will be post later, after tests. For first look Flux has better understanding of Riven's facial features and better at reproducing Nomo's artstyle. Here an example from training:
You don't have permission to view the spoiler content. Log in or register now.

For SDXL i forgot to write test promps, so ugly random images was generated)

I used only images from game (max resolution is 550x416).
 
  • Like
  • Red Heart
Reactions: StoryMan, Daggy, NaughtyOppaiDragon and 10 others
K

KGARrokh

Member
Oct 18, 2018
145
461
ComfyUI_00071_.png
You don't have permission to view the spoiler content. Log in or register now.
You don't have permission to view the spoiler content. Log in or register now.

Well model for Flux 1D is mostly dissapointing, i think it's not enough trained) Prompt understanding not so good, Pony is somehow better. I bet problems with promts caused due to my desicion to train with just a single trigger word in description "NN_R" and lack of experience.

What i can't understand is why Flux loves to draw nipples so frequent (most of time the not looks ok)... This is annoying because Flux doesn't support negative prompt.

I can't prepare dataset for training with full description, because of my low level of knowledge of english) I know just enough to read math and programming book, but not poems or other books.
 
  • Like
  • Red Heart
  • Yay, new update!
Reactions: Magnelo, true_exorcist, Eprofesy and 3 others
D

Dripstein

New Member
Jul 17, 2020
12
77
Another mini-story I made

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

I also understand a little more on image generation. Before, I was oversampling and had an over reliance on the detailer, which, not only caused it to take longer to get a good image, but would also make it look less like a KOD style. I'm at a good middle ground right now, but, with everything, it can be better.
 
  • Like
  • Red Heart
  • Love
Reactions: riggety, jg2142, nemore00 and 23 others
Vyacheslav Grinko

Vyacheslav Grinko

Active Member
Nov 9, 2018
590
631
Dripstein said:
Another mini-story I made

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

I also understand a little more on image generation. Before, I was oversampling and had an over reliance on the detailer, which, not only caused it to take longer to get a good image, but would also make it look less like a KOD style. I'm at a good middle ground right now, but, with everything, it can be better.
Click to expand...
This is a masterpiece make more of these mini stories this is better than the actual game holy shit
 
  • Like
  • I just jizzed my pants
Reactions: Ariaduob 25, StoryMan, big_shl0m0 and 2 others
K

KGARrokh

Member
Oct 18, 2018
145
461
00037-3697999625.png 00065-113021701.png

Trained another Riven model.
You don't have permission to view the spoiler content. Log in or register now.
You don't have permission to view the spoiler content. Log in or register now.

-overfitted;
-low resolution source images = bad proportions,
also some source images has bad proportions;
-biased to anime style;

+better character understanding
+works with other styles)

You don't have permission to view the spoiler content. Log in or register now.

Model by itself is bad, because of low quality (resolution 542x412, almost twice lower that needed for SDXL, but enough for Flux) source images, but with support models it decent.
 
  • Like
Reactions: true_exorcist, TERES, Darcel Marvis and 4 others
You must log in or register to reply here.
Top Bottom
5.00 star(s) 1 Vote