It depends on how fast you burn through credits. I have like 60 accumulated by now, because I just don’t spend more than 5 per day. Not that I’m consciously trying to ration them; I just don’t have more ideas to try than that.
The absolute bare minimum, based on a couple of tests, is 10 GB RAM (acceptable, but you won’t be able to get high resolution or accuracy; you are really going to want at least 16), and as for number of cores/speed you need at least a 2080-Ti or 3080 or equivalent (it’s still sluggish). And if that is all you have you may as well spare yourself the electricity and waste heat and just use the free NightCafe version.
I don’t have access to a true GPU farm either (at some level they cost real money to buy or run, so you need to use real grant money of your own, or persuade somebody else to run your jobs). I was able to call in some favors at an industrial video lab to sneak in some runs in the middle of the night, but their workstations are not configured for “deep learning”.
About the same. I bought 200 credits ages ago and still have around 130 left between daily free credits and the (now fairly rare) achievement credits. I don’t bother with the social media credits. No regrets since I have gotten at least as much entertainment from this as from many $20 video games. I have another 60+ on my tablet via a non-registered account which is less about fighting the system and more about me being too lazy to go to my desktop sometimes.
I suppose it depends a lot on how you use the site. I throw ideas at it and then say “lol, Nightcafe AI, you so silly”. Other people try to use it as a legitimate tool and burn through a lot in evolving images which probably gets expensive (in credits, if not cash) in a hurry.
Asking The Stars for Help --Harlem Renaissance style.
I’ve been burning through credits because I’ve locked myself into this paradigm of evolutions. As a result, everyone else’s work is absolutely amazing. I’m basically throwing out random stuff without really concentrating on trying to shape anything to any vision - just seeing what ends up splatting on the canvas. And I usually run at least low resolution - thumbnail is too small.
Here’s my MO:
I start with an idea for a title - The Hanging Gardens of Babylon, A Desert Flower - etc.
Then I pick out four levels of modifiers:
1st mod - two random artists.
2nd mod - three random Descriptive terms and two random Classic terms.
3rd mod - three random artists
4th mod - some concrete terms defining what I thing I’m looking for in the piece.
Now it gets tricky.
Mod 1 and Mod 2 get weighted at 4 - these act as the subtle underpinnings of the piece.
For the first run, I set Mod 3 at 5 and Mod 4 at 10 and run Coherent
Second evolution - Mod 3 at 10 and Mod 4 at 5 and run Artistsic
If you see what I’m doing there - cranking the “Artistic Mods” and the Artistic run and the “Coherent Mods” on the Coherent run.
Third evolution - Mod 3 at 10 and Mod 4 at 5 and run Coherent
Fourth evolution - Mod 3 at 5 and Mod 4 at 10 and run Artistic
(cranking Artistic for Coherent run and vice versa)
Fifth and sixth evolutions - everything cranked, including the “Subtle Mods” and run Coherent and then Artistic respectively.
The final final is rarely the keeper. Here are some examples from along the way:
2 Views from the Kosciuszko Bridge-- Hudson River School style. 1st is artistic and second coherent.
I wonder if it is referencing a specific painting or image.
I almost never have success with Coherent but it’s probably decide most of my prompts are more abstract. Probably works better with landscapes and “Ruined post-apocalypse city” type stuff.
I have a 95% rate of regretting spending credits on evolutions since I almost always prefer the original afterward.
Me too! And yet I can’t help but see what will happen if. . …
“This picture of Deer Next To Abandoned Car made for a really cool background, a nice looking car and this shape looks like it’s supposed to be the deer. Let’s give it another run to finish…”
Deer turns into an eyeball, car grows three extra windshields, trees turn into swirly triangles
Were you running both of those with the same seed image? If so, it’s likely that the parts of the original image that it thought vaguely resembled water continued to vaguely resemble water, and likewise with the parts that resembled shore, skyline, and so on.
Certainly none of those look at all like what I see when I Google for views from that bridge, but I’m not familiar with the area, so I’m not sure.
And inspired by that, I decided to see if it was familiar with my area. So I tried “Terminal Tower Cleveland”.
https://creator.nightcafe.studio/creation/Wez21a3NmCiWGX070UsH
Aside from including two copies of the tower, that’s pretty good. Even some of the foreground buildings are recognizable.
No, just the prompt View from the Kosciuszko Bridge run first in artistic and evolved with coherent, I did make it seem like it came up with the same thing twice, but no. Sorry about being unclear.
Here’s an example of something that “burned through a lot of credits”.
I had a general vision in my mind of what I wanted—a large pumpkin-headed creature rising/looming over a field of pumpkins. Creating a pumpkin creature is very easy—“pumpkin yokai”, pumpkin monster", “pumpkin ghost”, pumpkin demon", probably even “pumpkin prom queen” would work. Creating a pumpkin patch or field full of pumpkins is also very easy. Creating a creature rising above a field of pumpkins? Not. Easy.
I tried all sorts of prompts (some all in one, some split into two or even three sets) with all sorts of names (Mark Ryden is good for bulbus heads; will Dan Witz create a crowd of pumpkins?). I finally gave up on pure text and started making simple drawings, but the drawings weren’t simple enough to work—I had to lower the number of pumpkins to two or three large ones because a bunch of small ones get erased, and had to keep the sky some shade of blue (even though I wanted a night scene) because with black the contrast was too low and the pumpkin creature’s body disappeared. (Those simpler images still needed to be ran at 3 or at least 2 credit settings to preserve enough detail.)
Today I made a 2-credit (unsatisfying) image using the “cloud photography” modifier that I decided to evolve (1-credit size) without the cloud. I got a “wow” result. I then duplicated that image at 3-credit size and it got even better, adding bare trees in the background. It wasn’t what I imagined. It wasn’t what I worked towards. But damn, it’s a great pumpkin.
Here are some examples of images I made during the quest. (Note that I said “some”, not “all”.) I like a lot of these, too.
Portrait:
Landscape:
(This, BTW, is what one of the simplified start images looks like.)
I think for that, I might make an image of the pumpkin patch, an image of the Great Pumpkin, collage them together in an ordinary, Artificial Stupidity image editor, and then run that through one evolution just to smooth out the seam. At least, I’m guessing that’d work.
The Great Pumpkin was never actually seen on It’s The Great Pumpkin, Charlie Brown. (Though a verson was featured in a Simpsons Treehouse of Horror episode.)
(eta, I see you said “make”, not “take”, as in, “use a pre-existing”.)
I evolved my “walking pumpkin”, replacing the second prompt set (the one describing the pumpkin patch) with “psychedelic art | blacklight | tijuana bible”.
Looks like a scene from my childhood when my 'buela was practicing Santeria. I am not joking.
Anyone else noticed changes in aspect ratio during image evolution? That series of evolutions that went through start image to cloud pumpkin to walking pumpkin to bad acid trip kept getting more widescreen all on its own. It is subtle between any two images, but builds up with time. It went
1.777 : 1
1.857 : 1
2.000 : 1
2.111 : 1
2.200 : 1
2.333 : 1
Minutes ago I got 2 credits and a “patron” badge for getting my first follower and 5 credits and a “local” badge for getting followers. I already had 19 followers.