I’m working on stabilizing the AI rendering, but it’s not perfect yet.

Can you do predictive trajectory and then a transition to the next rendering?

Wow, this is next gen

once you get the blending smooth it will be insane. It is already super cool.

Looks like the inconsistency is because the seed changes with every camera change, right?

That's amazing, nice work

Max Payne vibes super cool

you should not only generate the texture but also the light maps and surface maps and use ray tracing

HOW TF

This is great idea, good luck

thats sick

You said it runs on low end machines, does it run on a steam deck?

Really interesting idea. Is it deterministic?

This is super cool. Would love to have something like this for the tool I'm building!

That's amazing! Can you share some details of how you stabilize the AI rendering? I'm trying to do a similar thing, but not for a game engine.

very interesting!

Games that have a stylized universe like Into the Spiderverse or horror games like a remake of Dreadhalls or something would be awesome with this. Maybe a noire? 😊

do you run llm 60 times per second? would it be easier to run it once for mesh and cache?

I've been waiting to see something like this, if it's what I think it is. I can see this being the next big thing in how rendering works. Am I right in thinking that it's a full screen image generation projected onto geo? Do you think there's an advantage of this being done in realtime, rather all of the surface textures being generated offline? EG you could have a main source image for the overall look/colour balancing/etc, then those get used but upscaled to do close-ups of individual parts.

take the de_dust2 demo you did months ago, add basic online fps multiplayer and release it into the wild 😅

why not just reverse-texture map the output?

Well, actually it can be a feature, not a bug. Kind of a style of a game where everything is flickering, glitching and changing a bit.

Holy fuck this is so cool

Can you connect bigger models for higher quality and a bit more delay?

What’s the use case here? It’s basically for development time so you can just walk around and fill up missing textures?

This is the right next step imo. The 3D scene should help keep the AI rendering stable across frames.

same shit happens in my dreams, can't get to a stable environment

What is your leverage to handle stability at this frame rate ?

looks like a worse version of a game made with photo scans

You will have to make it deterministic, and very good luck with that.

How does the update system work? When camera position or orientation has exceeded a threshold? And how expensive is the AI rendering for a frame?

Rooting for you

Please open source soon. And tell me the specs of what you are running this on!