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

[DavidS877]

After subjecting my system to Josh's MATI stream today. I decided to let it have some Kiwi fun.

Spoiler: Comfy Kiwis

 

[PandaChai]

I'm trying to generate Pre-Raphaelite styled Ecclesia et Synagoga for my phone wallpaper and lock-screen (respectively) and there are some good bases, but nothing yet I'm truly happy with. Honestly I'm surprised Bing let me generate them at all, really.

Spoiler: Image Dump

 

[WelperHelper99]

My Saber in Star trek series

Spoiler: Big dump

I used so many prompts for these, don't ask lol

 

[indomitable snowman]

Inspired by recent events, coming to you from TOR.
Become ungovernable, my Kiwi chads.

Spoiler: images

 

[inception_state]

I'm trying to generate Pre-Raphaelite styled Ecclesia et Synagoga for my phone wallpaper and lock-screen (respectively) and there are some good bases, but nothing yet I'm truly happy with. Honestly I'm surprised Bing let me generate them at all, really.

Neat, mind sharing your prompt? From looking at Pre-Raphaelite examples in a quick Google image search, my guess is that it hasn't been trained on the style. Your best bet would probably be to make a style Pre-Raphaelite style LoRA for Stable Diffusion. Basically grab 50+ images, tag them, then train on them with eg. https://github.com/derrian-distro/LoRA_Easy_Training_Scripts or Kohya.

Or if you don't have a beefy GPU, you can try the CivitAI free trainer: https://civitai.com/articles/4/make-your-own-loras-easy-and-free

 

[whatever I feel like]

I know you've all seen Cat-Dog before, but have you seen Koi-Horse?

 

[PandaChai]

Neat, mind sharing your prompt?

Two side-by-side rectangular Pre-Raphaelite [or specific artists; John William Waterhouse, John Everett Millais, William Morris] style wallpaper panels of the female representations Ecclesia et Synagoga. Ecclesia is crowned and holds a chalice and cross-topped staff. Synagoga is blindfolded and holds a broken spear and two stone tablets.

 

[BirdUp]

men of arstotzka

 

[unexpected onions]

I hate it so much

 

[Baraadmirer]

View attachment 5576041

I hate it so much

Do I even want to know what your prompt was?

 

[AnOminous]

"Slobbermutt bukkake."

 

[whatever I feel like]

Lol, Bing finally sent me to the cuckshed for an hour. It didn't like the surfing big bustle pregnant business women, I assume.

 

[ScatmansWorld]

"Largest Dataset Powering AI Images Removed After Discovery of Child Sexual Abuse Material"

https://www.404media.co/laion-datasets-removed-stanford-csam-child-abuse/

https://archive.ph/jFs3P

Looks like the start of a big shitshow for AI generated images.

Here's the Stanford study the article was referring to:

https://purl.stanford.edu/kh752sm9123

 

[The Mass Shooter Ron Soye]

Looks like the start of a big shitshow for AI generated images.

A thousand images out of billions probably doesn't make a dent in the trained model. Also, from flipping to a random page of the PDF:

Dead links: LAION datasets do not include the actual images; instead, they include a link to the original image on the site from which it was scraped. Given that multiple years have elapsed between the time the content was scraped and processed, a large percentage of the URLs passed to PhotoDNA (≈30%) were reported as no longer being active. They may, however, have been used to train models before they were removed from their original URLs, and some likely continue to reside in versions of the datasets retrieved at earlier dates.

If the image was deleted at the source URL long before a model could be trained on it, then it obviously didn't make it in.

We'll see if Stable Diffusion 1.5 leads to liability or law enforcement concerns, but mostly we'll see it abandoned in favor of Stable Diffusion 2.1, SDXL 1.0, and later.

Child sex abuse images found in dataset training image generators, report says (archive)

 

[Neurotypical Mantis]

32 fucking images warning

Spoiler: HRRNNNGGG I WANNA GO TO THE FRUTIGER AERO ZONE

 

[Baraadmirer]

Someone created a Chris-chan artsyle LoRA for SD 1.5 to keep the legacy of Sonichu going. There is also an SDXL version.

 

[Steve Harwell's ghost]

Recently on hacker news and info wars they showed the ant-White code in large llms. I have been trying for about a week with no luck to see this code in various llm's I have stored locally, to no avail. I've tried with vs studio, ida, and a few other programs. Does anyone have some info or hints on where to start? I've been working for about a year trying to fine tune my own local stuff. I'm not trying to create a gpt4 or anything, I'm not that stupid. My first problem is what I call "Lovecraft's cat" not to be edgy but as a matter of principle. I want to develop something that is exactly as described a machine that can pull info ie; ask a question from provided reference material then get the proper response. I have a lot of pdf's regarding various engineering topics so it would be nice to pull accurate info for a baseline then double check as I build a system (like a water heater).

Edit: I'm on gpt4all and would like to stick with it as I have been working with it the longest, I tried Kobold but with the recent update I have found it unusable.

 

[Mr.Miyagi]

I want to develop something that is exactly as described a machine that can pull info ie; ask a question from provided reference material then get the proper response. I have a lot of pdf's regarding various engineering topics so it would be nice to pull accurate info for a baseline then double check as I build a system (like a water heater).

From what I've heard, using something like LlamaIndex for Retrieval-Augmented Generation may be the way to go. I haven't tried it myself, so I can't offer much advice on setting it up for your local modes/frontends.

You may want to also ask in the ChatGPT thread, since this one is mostly about AI-generated artwork and models like Stable Diffusion.

 

[Balr0g]

A bit late but Merry christmas to all of you Kiwis

 

[whatever I feel like]

Praise the Lord, our Christ is born!