DarkViperAU's AI Upscale Turns a GTA V Run Into Chaos
An experimental AI upscaler promised a sharper Los Santos. What it delivered, according to DarkViperAU, was shinier faces, worse frame rates, and a speedrun abandoned by the third mission.
Somewhere between the promise of a sharper Los Santos and the reality of Michael De Santa's face sliding between three different ages mid-cutscene, you begin to understand why the prestige remaster is a craft and not a slider.
That, at least, is the shape of the experiment DarkViperAU describes running on Grand Theft Auto V Enhanced, as laid out on the channel's own upload: an experimental DLSS 5 setup layered over the game's existing visuals, toggled on and off through missions and cutscenes, with the stated goal of a full speedrun somewhere on the other side of it. The plan, per the channel, did not survive contact with the actual footage. By the time the mission Three's Company arrived, what had been billed as a speedrun had become, in the creator's own telling, a chaotic graphics experiment, and a decision to call it there rather than push on.
The interesting part is not that an upscaler produces uneven results on a decade-old character model; anyone who has watched a filter fight with pre-baked lighting could have told you that. It is what DarkViperAU reports the inconsistency actually looked like: Franklin, Trevor and the rest of the cast shifting in apparent age and shine between shots, occasionally landing on something the channel calls impressively realistic and just as often landing on something that looks like nobody in particular. Add input lag, a frame rate that reportedly dropped as low as one frame per second once the processing was pushed hard enough, and the unfamiliar quirks of the Enhanced build layered on top of rusty execution, and you have, by the creator's own account, less a speedrun than a stress test that happened to be wearing a speedrun's clothes.
One does not need to relitigate the entire argument about whether photorealism is worth chasing to notice what this experiment actually proves, which is that bolting machine-generated detail onto an art style built for something else is not the same discipline as building a remaster around that style on purpose; a studio that spends years matching lighting, geometry and performance to a single vision tends not to produce a protagonist who ages a decade between two camera cuts. DarkViperAU treats the result as a curiosity worth documenting rather than a fix worth keeping, and to judge by the footage described, that is the correct call. The video, as detailed at DarkViperAU, ends with the creator simply deciding enough is enough, which may be the most honest verdict any AI upscaler has earned this year.
