Hmm… do you mean a red herring about the method that might have been used to generate the image?
Really great stuff there!
I really like the entrances to the pyramids, the cemeteries, the jungle paths, and the caves - those are really nailing the “adventure game location” look for me.
The pirate towns are good too, some of the earlier ones are maybe a bit too busy/complex looking?
The last bunch of pirate towns you posted seemed to come out a lot better though, as there is not as much crammed into them.
What prompt are you using for the style, “SierraLucas”?
And what AI program/site?
I used your method. went to leonardo.ai, created a new model, trained with exactly these pictures:
https://drive.google.com/drive/folders/1IihCbgSKLXqZ8XTj_kDAEblLxfwPArq8?usp=sharing
this basically uses DreamBooth under the hood, to create a new checkpoint based on the stablediffusion 1.5 checkpoint.
I probably should have added borders to make them square, because this way I think stablediffusion cropped them automatically to be square.
and then I gave prompts like these:
pirate town at night, with shops, with palms
a pirate town at daytime, with shops, with palms
the exterior of a pyramid with a huge closed gate and two statues on both sides
a pirate cemetery at night, with tombs, trees
a path in the jungle with a parrot and other birds
those have “alchemy” disabled. it disabled automatically after a while. (free version)
1 Like
Excellent! I had been trying to find some original Sierra artwork (King’s Quest 6 in particular), but didn’t find any yesterday - those are really nice clear scenes for the AI to work with.
You put some of the MI2 art in there, did you choose those particular ones for specific reasons?
Also, I think Leonardo.ai only lets you train one model per free user (I had to keep signing up with different email addresses to try things out, haha)…
I tried deleting a model to see if I’d then be able to train another one using the same login/user, but it didn’t let me, it just deleted that model and I couldn’t retrieve it and had to make a new user.
For Monkey2, I picked the ones I liked the most.
Same for Sierra. I have many more, found somewhere on the internet. I can put them on google drive if you need them.
1 Like
If you have time, yeah, that would be great!
Would be useful for others too, if anyone wants to create a dataset from them.
It’s 2 GB. Fortunately I found the url where I downloaded it: ESRGAN Upscales - Google Drive
and to browse them, here: https://kq5upscale.neocities.org/ (not sure if they are all of them)
1 Like
You know what’s weird? With this style, I don’t think the pixelation improves things.
So my theory about pixel art being needed is imperfect 
With the proper brushwork, and proper texturing/graining, hi res is fine. In theory. Maybe it’s just practically impossible to obtain by humans. And in fact, all these images are the result of AI upscaling from pixel art. That’s not by chance: notice the grain, like a dithering:
2 Likes
I like both, but it will depend on how it looks with character sprites on it as well…
For me, pixel art blends the characters into the scene nicely, like they’re really there interacting with the background… while with hi res character sprites, they tend to look like they’re floating on top of the background.
Would be interesting to see if AI can make good hi res characters though, maybe it will work.
2 Likes
I haven’t read the whole thread in detail, so maybe this is obsolete, but have you and @seguso tried Stable Diffusion? You can use it offline and there are GUIs like Automatic1111.
1 Like
I think the site we’re using, Leonardo.ai, uses Stable Diffusion somehow…
I think Seguso previously downloaded the actual Stable Diffusion and Automatic1111, I haven’t tried that yet.
1 Like
I’ve tried installing stable diffusion and automatic1111, but it is very slow on my machine, and the default model does not give good results for me. I tried training a new model with DreamBooth but my card does not seem to have enough memory.
1 Like
To do list:
-Teach AI to come up with puzzles 
-Teach AI to make background art in “hand painted and scanned in” style
-Teach AI to make other art assets, sprites, inventory items, maps, etc.
-Learn a program like Adventure Creator or Adventure Game Studio
-Make up story with ChatGPT
-Write dialogue with ChatGPT
2 Likes
This last image reminds me of Laura Bow 2, inside the museum entrance.
That IS the bottom of the sphinx head. A.I. plagiarism?
It’s exactly that one. I’ve used it as an input to train the AI
that post contains two input images, not output. (which in turn were the output of upscaling)
1 Like
Oh, I see now! Heh.
1 Like