Microsoft says it has developed a new generative AI model that can produce “complex gameplay sequences.”
The Xbox maker’s Research Game Intelligence team debuted the World and Human Action Model (WHAM), which it has nicknamed Muse, earlier today and described it as a “generative AI model of a video game that can generate game visuals, controller actions, or both.”
It’s not for game developers, it’s to convince capital that they should keep funding Microsoft’s data center expansions. Keep funding the methane companies that provide the power for them.
Muse is currently capable of generating visuals at a resolution of 300x180 pixels
“Please, just give us more money so we can hyperscale enough for this to be useful” lmao
What would be significantly more useful for game developers is AI generated 3D models or something they actually ask for.
A stable diffusion plug-in or something to export 3d models for blender would be awesome. I’ve been waiting for something like that.
You can literally just generate .obj files from AI and have been able to for like a year.
They open in Blender no problem, but you would want to retopo, rig, and texture them for any real use.
Very much related:
LLaMA-Mesh by nv-tlabs on Github | Unifying 3D Mesh Generation with Language Models. Create 3D meshes by chatting.
meshgen by huggingface on Github | A blender addon for generating meshes with AI. This initial release contains a minimal integration of LLaMA-Mesh in Blender.
Something that would actually help you? Lol. No. We can’t afford that.
So they are marketing this to the Sinclair ZX Spectrum fans? It seems powerful enough, we only use 256x192 with 16 colors (two of which are black).
Good job, Microsoft!
I suppose it would be useful for low poly assets but I’d probably just make them myself because I can’t imagine the AI would do exactly what I wanted
All hype, No bite.
Okay but these models are actually pretty interesting when you dive in. Useful yet? Not by a long shot (I’m guessing decades, not years of work). But interesting nonetheless.
it wont
I just want an ai model that categorises player types and explains to me what and why makes certain types like some mechanics. It’s not useful in the least to generate visuals
What? I read that three times and it still doesn’t make sense.
In WoT game there was such a category of mods called “оленеметр”, literally “deer-o-meter”, where “deer” is a word for, ahem, the worst kind of “noob”.
I suppose that’s what they mean - a tool that analyzes players in a match by their previous games and stats and shows some colors and icons indicating that this one has good reaction, this one is likely to become erratic when their team starts losing, and so on. Also suggests best course of action against them.
I suspect that this is also how they are weaponizing these things IRL.
In machine learning, that task is referred to as classification
Ah, yes, there’s also extrapolation here though.
When you make games one of the hardest part is figuring out what works best for gamers that don’t like the same things as you, if that makes sense. I feel like a model could help with things adjacent to that
