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

5.00 star(s) 1 Vote

KGARrokh

Member
Oct 18, 2018
145
461
Could you by any chance help me and explain how to properly prompt different characters? How to group specific features between for example 2 girls (ex. prompt specifically Sabia to have purple hair and Neve to have white hair in the same image) I've seen a lot of 'methods' like using brackets or 'BREAK' but neither seem to work properly
Current generation of checkpoints is struggle with task) This also true for LoRA's.

All i can recommend:

1) Use special extensions (region prompter, masking).
https://f95zone.to/threads/kingdom-...-future-plans-healthcare.211283/post-14305546

2) Also you can try generate characters separately (1 character per image) with plain white/black background and then try to compose them in one image via photoshop or similar application. But method above is much simplier.

3) Wait for next generation of checkpoints, maybe they perform better at this task.
 
  • Red Heart
Reactions: Magnelo

Great730

Member
Oct 3, 2019
147
179
View attachment 3854596
Posted 2 new versions of KOD model. Added 1 more character: Jasmine) This time i used higher resolution for training, but result is still crap.
Also i noticed that earlyer epochs produses better results for some characters and bad for another, same true for later epochs, but later better understands character related details. Maply is good example of what i saying)
I think this is my last multi-concept model for KOD. Next models would be dedicated to characters (1 model per character).

Overtrained. Better at details.
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.

Undertrained. Better at reprodusing NOMO style, except for catgirls.
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.

No well trained model(

Waiting your feedback)

P.S. Now i can test all styles again:KEK:
How about making a bad ending with sabia and jasmine after enslaving wehlis?
 

KGARrokh

Member
Oct 18, 2018
145
461
image.png 1722682832017.png
New awesome model is out, called FLUX 1. But model is censored.
You can play with this model here:

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.

Model size is ~ 23 Gb) Bad hands, deformed bodies are included, but rare. Eyes generation is improved.
Also available for ComfyUI.
You don't have permission to view the spoiler content. Log in or register now.
 

KGARrokh

Member
Oct 18, 2018
145
461
Does anyone know any specific models that may look similar to those pictures? (using AustismMix SDXL) (I'll delete this post soon to avoid spam with off-topic content)
This looks close:

2.5d, no outline, soft shaded.
 
Last edited:
  • Red Heart
Reactions: Magnelo

..a..

Newbie
Sep 4, 2016
16
38
Anybody know workflow to fix finger or hand problems?
Sometimes I get good picture but keep generating too many finger or deformed hand
 

KGARrokh

Member
Oct 18, 2018
145
461
Anybody know workflow to fix finger or hand problems?
Sometimes I get good picture but keep generating too many finger or deformed hand
Adetailer, Inpaint, ControlNet, HiRes Fix can sometimes fix the problem, but in general for PDXL 6, there is no good solution for this problem. Even Flux Pro struggle with this problem, but less.
 
  • Like
Reactions: ..a..

Mamaragan

Member
Jun 9, 2019
344
1,335
Installed A1111 v1.10.1, looks like no difference from v1.9.4)
View attachment 3934984 View attachment 3934985

Also i tested ComfyUI, its great, but if you want save metadata... its overcomplicated, embarrassingly overcomplicated(
Actually, A1111 is faster and better with the update. I don't know what graphic card you've got, but with a 3070, using arguments --xformers --medvram --no-half-vae, it's actually fast. Even when using shit like Hires fix, controlnet and Adetailer.

Comfy... well... it's wayyy overwhelming. IMO it's only for seasonsed user who want to have access to cutting egde AI tech.

For the large majority, a1111 is enough. A real shame ForgeUI is depreciated.

00027-3040012131.jpg 00038-4277401822.jpg 00111-1763595189.jpg
 

KGARrokh

Member
Oct 18, 2018
145
461
Actually, A1111 is faster and better with the update. I don't know what graphic card you've got, but with a 3070, using arguments --xformers --medvram --no-half-vae, it's actually fast. Even when using shit like Hires fix, controlnet and Adetailer.

Comfy... well... it's wayyy overwhelming. IMO it's only for seasonsed user who want to have access to cutting egde AI tech.

For the large majority, a1111 is enough. A real shame ForgeUI is depreciated.

View attachment 3936069 View attachment 3936085 View attachment 3936086
RTX 4080 Super 16Gb with --xformers only. For me there is no singnificant difference.


Training models for Flux is now available on civitai with min cost of 2000 buzz. Not so much compared to image generation costs)
My GPU is capable to run Flux 1 [Dev], so i think i give it try and train character LoRA with Flux. But base model is still censoured(
Maybe next Pony model would be based on Flux.

Flux supported by ComfyUI and surprise new ForgeUI version)
 
5.00 star(s) 1 Vote