- Joined
- Dec 17, 2019
Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump
- Thread starter Near
- Start date
-
🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
- Joined
- Nov 21, 2020
My horrible SDXL workflow takes 90 seconds on a 3070 and 80 seconds on a 12GB 4060Ti
But I do "special" workflows.
50 steps for the image at standard resolution, then upscale and 20 steps upscaled.
- Joined
- Aug 28, 2018
That's overkill, you can do 25-30 with DPM++ 3M SDE Karras and then 10-15 upscaling steps and I doubt you will notice a quality difference.
- Joined
- Jun 2, 2020
- Joined
- May 2, 2021
I used bing AI with all of these. Just used a few prompts for each. Generally if I want a more cartoony style, I put in prompts like “drawn style”. Or if I want a realistic style, I just don’t put in the “drawn” prompt. Also, if I put in prompts that are possibly unrealistic, the picture will be drawn automatically. The rest are self-explanatory really.
- Joined
- Jun 2, 2020
- Joined
- Nov 14, 2022
- Joined
- Jan 31, 2021
Was trying to get album cover art for a digital-download single using some lyrics. Was not expecting these.
Last edited:
- Joined
- Aug 5, 2018
Monster iguana relaxing on a beach.
- Joined
- Apr 6, 2023
Started trying out the Automatic1111 ui. Playing around with img2img and making it look sort of like a painting. I haven't figured out how to upscale very well yet so these are small, but I like the results.
- Joined
- Dec 6, 2020
If you're looking to upscale AUTOMATIC1111 lets you do that from the Extras tab. I generally stick with R-ESRGAN 4x+.
- Joined
- Dec 12, 2015
You can also upscale through the img2img process; use a lower denoising strength (0.3 to 0.4) to preserve the original image but also fill in the new details visible from upscaling.
- Joined
- Aug 7, 2021
Bing really doesn't like a POV image, I wanted to churn out the night sky as if you're laying down and looking at it through the treetops but drawn in anime-esque fashion.
...it gave me 2B. Does it just ping the moon for Nier?
...it gave me 2B. Does it just ping the moon for Nier?
- Joined
- Feb 28, 2021
I'm impressed with the first London one because normally when you give it a city prompt it fucks up the geography (and especially with London tends to make lots of deformed Big Bens) but it's obviously improved as that's an accurate depiction of the Palace of Westminster and Portcullis House
Either the new model is getting better at referencing geography with landmarks or that's an edited version of a photo it stole.
I'm wanting to get into that optical illusion thing -
View attachment sQSqYoNY9G6y2XtPW69bXm-970-80.jpg.webp
but it looks like I'll need to run it natively (unless I buy some expensive subscription to pincel) and my laptop isn't up for it.
- Joined
- Aug 5, 2018
Flock of majestic cosmic bats flying through space
- Joined
- Dec 12, 2022
I got more coming later, but this one was my favorite Saber x American Civil War crossover I made. I call it Saber at the Battle of Gettysburg, Colorized.
- Joined
- Dec 6, 2020
Looks like Google has a proprietary AI ( Archive ) ready to roll out to the public.
Mr.Miyagi
kiwifarms.net
- Joined
- Jul 5, 2022
Here's the direct link for anyone wanting to try it out. Need to sign in with Google. Mentions a daily limit on generations, but that's probably easy to get around. Quality seems alright, nothing incredible so far, but I haven't tested it too much. Seems prone to complaining about TOS on innocuous prompts, unsurprisingly.
- Joined
- Dec 12, 2015
If you give it a more open ended prompt it tends to complain about TOS violations and it also tends to do smaller batches (or remove images) for anything that potentially involves people. Have a cat eating pizza.
- Joined
- Dec 6, 2020
Seems like it's currently country-locked.