Posted By
Luca
on 2022-08-05
14:52:46
 |  Re: Do you have some background infos about "DVM Andrea Brown Drawn By AI"?
Thank you @George, you anticipated me by opening a thread about that picture before I could do, and have to say it's pretty nice to verify that ppl had easily appreciated the experimental nature behind that release.
In the last years, several tools involving a self-learning AI have seen the light, in my opinion this big boom is a straight derivation of the diffusion of cryptovalutes mining's factories nowadays reaching even the homecrafting level in certain contexts.
We played textual adventures created on the fly with AIs capable to invent a long story you will like because it's based on the user's tastes after many attempts; we used AI's who translates on the fly some obscure Korean RPGs, or subbed some African series, and the result is still decent at now.
The new frontier opened a couple of years ago, leads to a creative application for an AI, from music to graphics and beyond, and as Csabo pointed out, today we have various implementations for AI's image generators. Among those, Midjourneyin its own V3 routines, actually looks like a further step toward this field of application, as the results versus the time spent around it are quite surprising!
Midjourney is Discord driven: after you joined it, you can prompt in a three-parts string: an optional existing reference picture, a verbose description of the subjects assigned to be drawn, and some eventual parameters to be added. Once entered, the AI begins to draw a quarted of possible pictures, from where you can pick up one to either being further evolved in 4 new proposed ones, or upscaled in a definitive format where new particulars would be added and better defined.
The prompt I used for DVM Andrea Brown Drawn By AI is: A smiling happy Afroamerican young girl with afro haircut wearing a tshirt and big headphones around her neck. Toon-styled snakes behind. Iris gradient in the background. Bright vibrant colors --s 15000 --q 2. This produced the following quartet.
I've chosen the last one, in the bottom right corner, and the AI upscaled it to produce this picture.
After a brief time of training to learn how to use the parameters in a decent way (everything's clear on their info pages btw), I spent...mmm...say 17 minutes. 17 minutes all included from the time I started the AI to the time I loaded the result on a Plus/4. Seventeen minutes, yeah!
Beside that picture, I did another experiment: I've taken one title of an existent text only adventure game, and then I've driven the AI in order to obtain several bitmaps of precise shape and resolution, to promptly convert and save MC bitmaps to be used for a possible graphics to accompain the game. In about 90 minutes I had 4-5 evocative bitmaps ready to be added.
There's one thing I am sure of: if I could have these tools in the 8bit golden era, someone should have to call an ambulance for me and push me off from the screen because I'd eventually never stopped to play with it and converge all my efforts to weird Plus/4 projects to cover  You can literally have your own Level 9 styled adventure game one per week!
|
