Anyone noticed that i was gone? :D There is a post about windows deciding not to work, backups when in crypto space, some surprises with Brave backups... that will probably never be written :D
So back to some art fun.
Disco Diffusion is an AI text-to-image app. You give it text and it tries to make an image from it. And it is fun to watch what it can do. There are some surprising results.
But for me there is an option in Disco Diffusion that is even more fun/frustrating :) Giving the app an initial image, setting a starting point and then waiting to see will the result be cool or complete garbage.
If you followed development of 3D animation (or just watched movies in the last 30 years) you probably noticed that the last thing creators try to do is make a realistic human characters. Because it is hard :)
From my experiments with Disco Diffusion it is the same thing. I was actually surprised how well it understands what is what in the photo. But still it does some crazy things to it.
There is a lot of try and error in the process. Especially with picking the photo. Overall face strait to the camera works best, but even then not all photos work. I was kinda lucky with this one as i picked it really early in the try and error process. As if not maybe i would just gave up. From my testing just head portraits don't work at all.
Then there is the text you enter into Disco Diffusion. Some things work great and some don't work at all. In these edits Steampunk (steampunk glasses), tribal warrior worked nice. Anything with sci-fi, alien, robot, android looked really bad.
All of these photos are mixes of several results from Disco Diffusion and original photo. As almost every version has something interesting in it, but the rest of the photo is just all over the place :D
I found that this photo also works ok with "elf wizard" combination.
And this is the original photo used.
Excellent work with Disco! It's cool to see other folks on Hive starting to get into AI art! Have you played around with any of the other text-to-image google colabs around?
Didn't do a lot of research, i kinda stumbled upon Disco and play with it from time to time. mostly with bad results :D
And i seen DALL-E (mini is just for fun) but there is some waiting list to get access so i just gave up.
Any recommendations?
I'm still on the Dalle waiting list, so no tips there, but I would recommend joining the Disco Diffusion discord server, as well as checking out the AI art scene on Twitter :)
That's cool. I'm using NightCafe Studio. But a few days ago I found this DALL-E mini it's said that's the second generation of Google's AI.
will have to check NightCafe. I tried Dall-e mini, it is fun but you have no options to play with. It kinda became a funny picture meme generator :D
If they only added an option for people to pick the result they like it would be a AI trainer :)
DALL-E is primitive. Dream Wombo is much better. https://app.wombo.art/
Thanks for link, will research :)
Sign up on it - then the site will save all your work, as long as you clicked publish. You can download the stuff, and I crop it in Photoshop and even enlarge the resolution size from 96px to 200px, it comes out pretty good. Sometimes a bit of smart sharpen helps and increasing the saturation. Here is one piece I worked over that way in Photoshop. I cant remember my sentence anymore, other than it had witches and the devil in it. The style I used was Surrealism.
Blocksberg
DALLE-E mini isn't the same thing as DALL-E 2, it's an attempt at recreating DALL-E 2 (which is invite only right now), but open-source. It's a lot of fun, but it's definitely not the same level of insane quality as DALL-E 2 :)
That's an I teresting program. It sounds like it would be kind of fun to see what it comes up with. But I'm sure the novelty wears off pretty quickly. To me the wizard version looks best. It looks the most intentional and less random then the others.
The main problem is the time it takes to make one of the variations. And google is limiting GPU that is willing to give you for free :D so it is a long process with mixed results. but it does have endless possibilities and when ever you think you had enough of it you get something cool 😀
wizard is simple but nice edit. other are a bit too much.
FWIW, Google's premium options seem to be fairly inexpensive for what kind of GPU horsepower they grant you.
seen some people talking about that pro and pro + options and prices, if you are doing something serious prices look ok.
I've been on Google for many years, at least a decade. Using mostly Drive for docs but I kept running out of storage space, more so because of my email, so I went the plus route and find it is worth it, the price of a cup of coffee for a month (at the level I am on).
For photos I got a Flickr Pro account, unlimited storage. It doesn't break the bank either.
That's probably by design. It just messes with you to keep you coming back so they can mine your data, until it thinks that your about to leave at which point it gives you something good to start the process over again...😂
I'm just joking but that's kind of what it sounds like to me, an algorithm built to keep you on site haha 😆
i mean we are teaching AI with google computers so i am sure they are using it for something. Never got into how colablaboratory works, but i am sure they are not giving it for free just because 😂
I don't even know what colablaboratory is. Do I want to know?
Lol
my understanding of it is that google gives you their cool computers with cool GPU units so you can run AI training for scripts that you written (or others did). Why no idea :D
but i am not sure because i always go backward when trying something, first make it work with most simple tutorial and then click buttons to see what will happen :)
Okay that's interesting. Yeah I'm sure they have their reasons for doing it though. They probably own anything created on the device I'm guessing. Just a guess though 🤷♂️
I really like using DiscoDiff I find the results really amazing your work reflects it really well congratulations
thanks. it is a lot of try and error, some results are fun
Same for me a lot of things I don't keep and some really surprising results that I keep preciously
and i am still not really sure how to get more detail in the images. is it just more steps.
Really Awesome Editing 😍 U r amazing @bil.prag
thanks
you changed your profile picture, almost did not recognize you :D