AI Image Generation

I've been restoring my old BSP libraries to life again. This is a map compiler for old quake style map editors. I primarily use quake army knife.

Anyway, one of the best ways to jazz up a map is with a nice skybox. These are 6 sided cubemaps that any face marked as "sky" projects into. They used to almost always be made in Bryce. An ancient tool some of you may remember.

Now I'm attempting to use stable diffusion. My first attempts using "panoramic forest skybox" with mostly default settings got this sort of thing:


Not very useful so I switched to mountains:

Got alot better results. Occasionally it would split vertically thusly:

Then I tried a superwide format that I could easily wrap onto 4 cubemap faces 256x4:

These got really vague and spooky and arty. I think the cause was I activated some automatic tiling checkbox.
It also seemed to want to put vertical bars as if looking out a window.

Eventually I switched from Euler A to DDIM and it makes some really nice scenery:


The real trick is going to be the top and bottom faces of the cubemaps. Bottom is rarely seen but it is nice to have.

Here's some "forest, looking straight up at the sky". I had to go 512x512 as 256 doesn't seem to be big enough for it to work with:

This vid has it figured out. There's a pluging + using monoscopic 360 vr seems to be the key.


This gets me results like:


So have to figure out how to convert that into a cubemap

found this:

That github thing is not too bad! I don't have it in game yet but these are the faces for a mountain set:


Makes you want to mine some iron ore!

anyone keeping tabs on ai porn? Lots of advancements lately had really impressive results putting celebrity faces on some 5ways etc. you can obviously tell it’s fake but it’s waaay better than it used to be

Here's what the above mountain box looks like in my map build tool in a semibroken map:

Here are some experiments with various models with the prompt "videogame mmo fantasy cover art"


This one I did properly with ffmpeg. The previous I was lazy and just used an irfanview slideshow captured with obs

Good site for grabbing models Civitai | Stable Diffusion models, embeddings, hypernetworks and more

Might try the pose detection stuff. Curious if I can extract it in a usable form, like bvh or bip or something

Hmm the pose model is 5 gigs. Probably too much for my poor old 980 4gig card.

We helped Rob learn inpainting. The mouth worked well but he had a really hard time with the hand.