A quick experiment with Grok highlights another problem: AI output is inevitably guided by the prompt used to generate it. The ability to easily influence facial features in the "enhanced" image is another more reason using generative AI to enhance images is not a viable way to ID unknown people.

Similar "enhancements" appeared following the killing of Alex Pretti. One popular AI-enhanced video includes several frames where the gun taken from Pretti appears to fire at the ground while in an agent’s hand, spewing wildly unrealistic flames. This event appears nowhere in the real video.

Other images circulated after Pretti’s killing include attempts to increase resolution with AI. As with other “enhancements”, this does not work; as the AI models are simply filling in missing detail with info derived from their training data rather than anything to do with the actual shooting.

More detail in this article: www.conspirator0.com/p/computer-e...