A GTA 4 video created using an AI-based video renderer has raised a lot of dust on Reddit. In the future, will we continue to use pixels or rays to create games, or will we move to real-time neural rendering, as Nvidia sees as the future? The GTA 4 AI video has certainly attracted interest.
Don't miss our articles, like us on Facebook too!
I'm following you!
A modified version based on artificial intelligence for GTA 4
Reddit user indiegameplus posted a video that he claims was created with the help of the Video2Video AI app, which shows the possibility of remastering GTA 4. The user wrote: “GTA 4 with Runway's photorealistic filter – shows an exciting future for remastering.” According to the creator, he created the video using Runway's latest Gen-3 Alpha model. With Gen-3 Alpha, you can create videos based on text instructions, but you can also create generative visual effects (GVFX) similar to live footage from still images, or even create new videos from video clips. Among other things, Runway promises realistic human faces.
It's about this video:
On Reddit, the creator wrote: “I've tried both Gen3 and Gen3 Alpha Vid2Vid – these shots are all from Gen3 as they were clearer and more consistent, but other than that the Alpha Gens are very impressive and took about 15-20 seconds to produce. The Alpha makes me very curious about the real – time aspect of the future – They keep increasing the speed and I know Nvidia is working on neural filters like this which can almost be considered the next step for DLSS and DLAA – and developing their RTX-Remix technology, I wouldn't be surprised if we have playable versions next year, they probably won't be perfect, But it could be the next level.”
Video analysis
The video first shows a red sports car driving down a well-lit stretch of road reminiscent of Times Square, mainly due to the video's advertisements. The main character then wanders into the TPS show, enters a restaurant, and attacks passersby – including stealing classic cars. Finally, we can see an unusual trick with the motorcycle. The atmosphere of GTA is captured, although the process is not always entirely logical.
the future
It will be interesting to see how the neural display continues to evolve in real time. Nvidia has been working on neural features like “Radial Caching” and “Radial Texture Compression” for some time, which could be added to feature-specific DLSS in the future to replace much of the rendering process with AI. In September 2023, it was confirmed that DLSS and AI-based image reconstruction could one day replace classical rendering as we know it today. Of course, that's not certain yet – until then, videos like the ones above can only be considered interesting games.
(This article originally appeared on the German website pcgames.de.)
