Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
Here's a question for those of you using SD more, what's a good general-purpose model that works in Automatic1111's interface? I'm looking for something that'll pretty much generate anything, SFW or NSFW, not necessarily specially trained for anything in particular.
 
Here's a question for those of you using SD more, what's a good general-purpose model that works in Automatic1111's interface? I'm looking for something that'll pretty much generate anything, SFW or NSFW, not necessarily specially trained for anything in particular.
At this point the two have really diverged. SDXL is really the best for SFW. While PonyXL is the best for NSFW easily. You are still going to have problems with text.
 
  • Thunk-Provoking
Reactions: Vecr and anustart76
I went back to flux after a few months of not doing anything and its gotten really good. All on its own it made this great tancho without even any specific prompting.
 

Attachments

  • japanese_woodprint_of_jumping_koi_in_winter.png
    japanese_woodprint_of_jumping_koi_in_winter.png
    1 MB · Views: 39
Here's a question for those of you using SD more, what's a good general-purpose model that works in Automatic1111's interface? I'm looking for something that'll pretty much generate anything, SFW or NSFW, not necessarily specially trained for anything in particular.
Right now it's Flux leading the pack for general purpose image generation, but it isn't great with NSFW, so Pony still has a large niche for 2d NSFW and some people are still using SDXL variants for realistic NSFW. That's just my perception from casually skimming CivitAI.

Unfortunately A1111 does not have Flux support, so you'd have to switch to Forge (similar interface to A1111, I think it's a fork?) or use Comfy.
 
  • Thunk-Provoking
Reactions: anustart76
Right now it's Flux leading the pack for general purpose image generation, but it isn't great with NSFW, so Pony still has a large niche for 2d NSFW and some people are still using SDXL variants for realistic NSFW. That's just my perception from casually skimming CivitAI.

Unfortunately A1111 does not have Flux support, so you'd have to switch to Forge (similar interface to A1111, I think it's a fork?) or use Comfy.
my prediction is shit is really gonna hit the fan legally if and when flux becomes able to do NSFW stuff consistently. up until this point there have been NSFW/ porn A.I images of celebrities but you could always tell there was something off about them, like they were people who look similar to the celebrity but it wasn't exact and you could clearly tell it was A.I. but flux is getting loras that are so accurate, so indistinguishable from just a straight up photo that it's gonna cause a major backlash. especially since pretty much every celebrity on earth (and potentially anyone in general) is going to have an incredibly accurate model just waiting for some gooner to spam super realistic porn of them all over the internet.

imagine the backlash those ai images that got spread around on twitter last year of Taylor Swift getting fucked on a football field but 10x worse.

00018-884986492.png

00047-3788894726.png

00065-428087378.png

00095-3053058848.png
 
Here's a question for those of you using SD more, what's a good general-purpose model that works in Automatic1111's interface? I'm looking for something that'll pretty much generate anything, SFW or NSFW, not necessarily specially trained for anything in particular.
As someone who recently moved on to using Forge, which is also a UI similar to A1111, I suggest using that instead as it generates images a little faster and loading different models seems near instantaneous. Dreamshaper does pretty well for general images, but the images it produces have a dreamlike quality per the name.
 
I was watching (more like listening) to a 40K lore video and it ended with the following indie folk song set to highly contrasting (i.e. violent) animation from that universe:
After failing to find the associated band I suspected it was AI and it is:
It even has a little chuckling in the audio before the recording proper starts. I'm really quite impressed.
So, I assume the lyrics were written using AI as well, then? I can see including it with violent imagery, if only to increase the total sense of unease…
 
Cool. I liked the part where the guy’s head was set on fire.
That came from the (human-generated) animation The Faith:
The animation is a bit clunky compared to the most recent work from Gabriel Christtiane, The Awakening, and the heavy Slavic accents and wording are a bit off but it's still really quite good in my opinion.
 
  • Informative
Reactions: Vecr
been playing around with flux in comfyUI and I'm still blown away by the quality of the images it produces. I made a model of my favorite old Irish singer and it's incredible that even with a dataset with 90% low quality photo it can somehow produce high quality images that are not only consistent but actually look like her. all while the Lora is only 18mb where SDxl Loras are 200+mb and don't produce nearly the quality.

ComfyUI_temp_xnvzx_00013_.pngComfyUI_temp_xnvzx_00017_.pngComfyUI_temp_xnvzx_00018_.pngComfyUI_temp_jcsos_00011_.pngComfyUI_temp_vxvux_00045_.pngComfyUI_temp_vxvux_00033_.png
 
Last edited:
been playing around with flux in comfyUI and I'm still blown away by the quality of the images it produces. I made a model of my favorite old Irish singer and it's incredible that even with a dataset with 90% low quality photo it can somehow produce high quality images that are not only consistent but actually look like her. all while the Lora is only 18mb where SDxl Loras are 200+mb and don't produce nearly the quality.

Why you wanna fuck da queen?
 
  • Thunk-Provoking
Reactions: Train Operator
Back