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

A few based around the original prompt: “A 30 year old man sits alone in a Serbian apartment eating pizza while browsing the internet on his desktop computer in a dark room lit by a warm light. He is wearing a cozy blanket that has cute cartoon orange and white hamsters on it. ”
IMG_3640.jpeg
IMG_3645.jpeg
IMG_3654.jpeg
IMG_3655.jpeg
IMG_3657.jpeg
IMG_3658.jpeg
IMG_3660.jpeg
IMG_3662.jpeg
The prompt varied depending on any other stuff I wanted to add or change.
 

Attachments

  • IMG_3661.jpeg
    IMG_3661.jpeg
    143.3 KB · Views: 29
  • IMG_3659.jpeg
    IMG_3659.jpeg
    142.1 KB · Views: 15
  • IMG_3656.jpeg
    IMG_3656.jpeg
    122.6 KB · Views: 15
  • IMG_3649.jpeg
    IMG_3649.jpeg
    108.6 KB · Views: 15
  • IMG_3648.jpeg
    IMG_3648.jpeg
    114.3 KB · Views: 17
  • IMG_3647.jpeg
    IMG_3647.jpeg
    134.4 KB · Views: 13
  • IMG_3643.jpeg
    IMG_3643.jpeg
    124.4 KB · Views: 12
  • IMG_3642.jpeg
    IMG_3642.jpeg
    108.3 KB · Views: 21
  • Like
  • Winner
  • Feels
Reactions: Bungdit Din, SauliNiinisto, The Jesus Lizard and 12 others
Shiawase said:
Sorry about your account. Due to the way this system works, I may be next. I'm not trying to push the boundaries or cause a scene, but I received two warnings of suspension without any explanation of what went wrong. Microsoft #1 anti-fun company.
Click to expand...

You guys may be playing more guinea pig than you realize. Microsoft recently mad some big announcements, where they showed off a bunch of of AI integrations into their core Microsoft 365 products (formerly Office 365). They have an "enterprise" version of Bing chat AI, added the ability to summarize emails and your work day, and added the ability for their existing automation tools to have the AI generate documents. Copilot was added to basically everything.

What I haven't seen yet is the image generation capabilities in all this. Right now they're sticking to text analysis, code generation, and document generation (which combines the previous 2 capabilities). So it's possible you're showing MS where the limitations need to be, so they can sanitize image generation enough to move it to the enterprise side and start charging for it.
 
  • Thunk-Provoking
  • Feels
  • Like
Reactions: Bungdit Din, Haftag, Ughubughughughughughghlug and 2 others
anustart76 said:
View attachment 5423709
Anyone tried this latest Nvidia driver update? Says it speeds up Stable Diffusion by up to 2x.
Click to expand...
Per discussion on Reddit:
It looks like it takes about 4-10 minutes per model, per resolution, per batch size to set up, requires a 2GB file for every model/resolution/batch size combination
Click to expand...

It does work for higher resolutions and SDXL, but there is a fair amount of upfront cost to set it up.
 
  • Informative
Reactions: The Mass Shooter Ron Soye and anustart76
I can't seem to find more info on fine-tuning prompts for bing. It seems to be really hit or miss and very arbitrary. For some reason, I'm told anything LGBT related seems to pass through?
 
  • Like
Reactions: Neo-Nazi Rich Evans
Is SDXL actually worth trying to use on an 8gb GPU or can you get better results with SD1.5 with a custom merge and some LoRas?
 
  • Thunk-Provoking
Reactions: Toolbox
Epoch said:
Is SDXL actually worth trying to use on an 8gb GPU or can you get better results with SD1.5 with a custom merge and some LoRas?
Click to expand...
I couldn't get it to run using A1111, but ComfyUI seems to work fine with SDXL and 8GB GPU, on Linux. Slowly, but fine.
 
davids877 said:
I couldn't get it to run using A1111, but ComfyUI seems to work fine with SDXL and 8GB GPU, on Linux. Slowly, but fine.
Click to expand...
I tried the leaked version on Comfy, but it was painfully slow (minutes to do a 512 image with the upscale to 1024) and I got similar or better looking results with 1.5, I just wonder if that has changed at all.

I also don't really like Comfy's UI for actual generation. Its nice for messing around, but actually just iterating on stuff I find it to be more annoying.
 
Epoch said:
I tried the leaked version on Comfy, but it was painfully slow (minutes to do a 512 image with the upscale to 1024) and I got similar or better looking results with 1.5, I just wonder if that has changed at all.

I also don't really like Comfy's UI for actual generation. Its nice for messing around, but actually just iterating on stuff I find it to be more annoying.
Click to expand...
It is a pain in the ass, but once it's setup it's not too bad. There's a few QOL things like adding the LoRA tag lookup for making iterating less "clicky".

But much of what I like to do with it is have it select a random checkpoint, with a random/wildcard prompt, with random LoRAs and let it run a bunch and see what pops out.

beautiful {sunny|cloudy|rainy|snowy} day {<lora:beach:0.8>at the beach|<lora:bigcity:0.9>in the city>} with a {red|yellow|green|blue} {pickup truck|<lora:exoticcars:0.8>Ferrari} ....

This is not that prompt:
SNEED_7.jpg
 
  • Like
  • Informative
Reactions: The Mass Shooter Ron Soye, Neo-Nazi Rich Evans and Toolbox
Can anyone tell me where I can do AI images for free and with good quality? My midjourney account got run out of credits and I don't know want to pay.
 
Screamer_2 said:

It can be sensative with blocking stuff though.
Click to expand...
Best I can do.

1697684630444.png

"sensitive"

Jean Arthur brandishing a dictionary at a person. The person is saying "sensative." Jean Arthur is saying to the person with the dictionary "Learn to SPELL!", animated
Click to expand...

Back to my Dall-E:

1697684682699.png1697684720629.png1697684753631.png1697684801580.png

Now I'm hungry.

Jean Arthur with a tray containing a meal of cheeseburger, fries, and drink in a cafeteria. colorized
Click to expand...

This is definitely giving "girl next door" vibes.
 
  • Feels
  • Like
Reactions: Bungdit Din, The Mass Shooter Ron Soye, Obi Wan Kenobi and 2 others
You must log in or register to reply here.
Back