AI digital art (Dall-E 2.0, Midjourney etc)

I quite like Adobe's take on this, or at least from what I understand of it.
Basically they let their AI run wild through their on stock image collection. ONLY their stock image collection. These are images voluntarily submitted by creators, who receive royalties for the 'use' of their work. The database is quite massive too, and will likely only get bigger.

I'm sure that policies and rates of compensation will also be adjusted over time - hopefully we can find good common ground.

 

Currently playing around with Intel AI playground with my new arc graphics card..

Wondering if there was an engineering related AI model I could use?

Trying to generate images of a distillery or milk production etc for blogs and beers etc for my business

 

Fun playing with various models locally. Learning positive and negative prompts, keyword weighting, etc. Using a 3060 Ti 8GB. Got a 4060 Ti 16GB on the way now.

 

the ai / deepfake x rated videos people are making with kling ai these days is kinda crazy too, i can see a lot of people getting very very angry very soon

 

Would be very interested to see what improvements that brings.

 

It arrived today and goes alright. Extra VRAM allowing me to generate higher res images and use different models.

I should have kept a copy of my prompts and seeds so I could regenerate as a comparison. Oh well, from now on.

 

Its embedded in the exif data. eg. for the first 2 images that I just looked at.

 

Sweet, thank you. Though, for some reason it seems that my Flux models aren't working. When they're selected it seems that the previous loaded model is actually used. So, the tattooed guy is generated from the "Model hash: 84d76a0328, Model: epicrealism_naturalSinRC1VAE" model.

 

JSmith

 

im still generating woodland nymphs

i used dall e 3 last night.

 

hahaha.







nymphs

 

my pick of pics for night cafe AI art generator (of nymphs)

 

beautiful ?
Serbian

 

This stuff is insane...



... and its just beginning in its technical capacity to fuck over a large portion of the connected citizens.

 

Oh wow...

https://www.instagram.com/reel/DEa-7Q_obGQ/

Edit: did some light stalking on the artist... came up with these links;

• https://www.inabsentia.it/gallery/the-intergalactic-museum-of-alien-oddities/
  
• https://fredrikjonsson-33401.medium.com/list/tools-and-assets-5821760e100c

The last link could be interesting for anyone interested in generating their own AI art - it seems he has an article on prompting... although its from 2023, it might still have some tips and tricks that is valid for todays engines.

 

Its funny.

I think one of the most interesting things about AI is what it says about what you want to generate. When I use it I have specific things I want.

Unfortunately some of the AI's aren't very good at creating totally weird requests. A lot struggle with my requests for weird robots. They are very good at generally generating women, so much so that sometimes it can't generate anything unless I request a woman in it.

They suck at lizards and reptiles and many animals. All my lizards look like iguanas, all my crocodiles look like alligators. Three quarters of my prompt is often spent on animals.











I quite like Meta AI ability to handle my weird requests. But my requests are so weird it doesn't use models that work well at generating realistic humans..

 

With AI you need to know how to prompt as each model or service will respond differently depending on their training. Some like natural language, others short descriptions, others just trigger words.

Overall you will get better results by downloading models and loras specific to what you want and running it all locally on your PC.

Its a bit of a learning curve but worth it.

 

I was doing this. But my RX6800 is pretty slow, and I was having issues with AMD hardware trying to patch it in to make it work.
I like using the free web based ones. It can help with composing a scene or terminology.
Sometimes I use latin names for animals as I get more scientifically accurate results.

I have a 5070Ti coming. So it will be much easier to play with AI locally. I wonder if I am just playing with AI if I should have gone with a XTX just for the memory.. I hate generating small images, even with upscaling I can create complex scenes or images..
That second image was done with local AI on stable diffusion.

I do like the offline models because you can get much grittier images. Most of the online tools generate very safe images, but they are almost like PG cartoons now.

 

I have 10gb and can generate 1.5k images then upscale to 3k fairly easily or double that if I sacrifice some details using a GAN. It does take over 5mins to complete the upscale though because of the lack of vram, yet still possible.

This is using Flux with a small section of the 2x upscale pasted for example.

 

I have played with upscalers a bit, but I find they change the tone of the work. Such as in your example, the upscale looks much younger. Still a great image.
It also annoys me because I have to stop firing off new images to focus on upscaling.

But I am going to redo my workflow with my new GPU.. Try some new stuff and settings.