AI [AI Art] - Show Us Your AI Skill *NO TEENS*

4.80 star(s) 6 Votes

Astcit

Member
Jan 14, 2021
443
845
It's both annoying and amusing in equal measures how I can prompt the most absurdly wild, perverted and obscene things, and yet something like this comes out:
ComfyUI_temp_nbbgp_00035_.png
Also, this new horse pussy lora is kinda good:
ComfyUI_temp_nbbgp_00024_.png ComfyUI_temp_cjrza_00018_.png ComfyUI_temp_nbbgp_00011_.png ComfyUI_temp_nbbgp_00017_.png

I see many people switching to ComfyUi. Did you know why? Can LoRa work's on it?
I, for one, think ComfyUI is more direct and with less happening under the hood that you can't directly control.
And yes, you can use Loras/Lycoris:
Untitled.png
 
Last edited:

Astcit

Member
Jan 14, 2021
443
845
What was the prompt? I find sometimes if you use too many words it might make things worse and confusing for SD, best keep it simple if you want something specific
It was only 12 words, too much gets plain ignored if you write a book in prompts.
It's just too dumb yet, it doesn't understands anything, it just translate words into search queries for image patterns.
Maybe 10 years from now it not only analyses words and sentences, but their actual meaning when put togheter.
 

devilkkw

Member
Mar 17, 2021
337
1,143
It's both annoying and amusing in equal measures how I can prompt the most absurdly wild, perverted and obscene things, and yet something like this comes out:
View attachment 3168391
Also, this new horse pussy lora is kinda good:
View attachment 3168392 View attachment 3168393 View attachment 3168394 View attachment 3168395


I, for one, think ComfyUI is more direct and with less happening under the hood that you can't directly control.
And yes, you can use Loras/Lycoris:
View attachment 3168398
Thank's for reply. Interesting. what is diferrence about strenght_model and clip? never seen option like that in a1111.
 

Astcit

Member
Jan 14, 2021
443
845
Thank's for reply. Interesting. what is diferrence about strenght_model and clip? never seen option like that in a1111.
I have no clue at all. There's no actual documentation on it, and the code is fucking soup.
I tried variations of it, but it never actually seems to make any factual difference apart from lowering or raising both model and clip weight togheter (setting them both to 0.8 or 0.2 acts exactly how A1111 does Lora weights, but making them different values does something I can never predict nor understand so far, it becomes wildly random results).

And that's a problem with ComfyUI, either small details like this one are never explained anywhere at all, or terms like "weight", "clip", "noise", "steps", etc, have different meanings and logic behind them than how A1111 does it, for example.

IMO, it's a standards issue. All of the clients, webuis, etc, all of them follow no actual standard. So people have to spend days tweaking and testing things around on any of them to catch glimpses of what something does and how it does it, and only then produce something actually worth the time invested. That's both good and bad in equal measures.
 

Astcit

Member
Jan 14, 2021
443
845
This new LCM stuff is VERY impressive. I went down from 30 sec 960x720px images to 5 fucking seconds! And I'm using a AMD card with DirectML, no CUDA or Xformers bullshitery (6900XT at stock clocks). Fucking magic. Hope this keeps getting optimized, it's very impressive already.
ComfyUI_temp_xberp_00125_.png ComfyUI_temp_xberp_00151_.png ComfyUI_temp_xberp_00154_.png

Workflow if anyone is interested:
Screenshot 2023-12-13 at 23-36-45 ComfyUI.png

And the gen times proof:
Untitled.png
 
4.80 star(s) 6 Votes