Remember the first popular deepfake in 2018? The BuzzFeed/Jordan Peele “Obama” demo was the moment a lot of people realized synthetic video could plausibly launder misinformation; but it was still a project.
In 2026 I estimate this would be ~99.6% cheaper and ~98% quicker to make.
In 2026, this minute is basically ~$1/min for the video plus ~$0.30/min for a cloned voice; and it’s no longer a multi-day training run; it’s closer to “generate/iterate” in an hour (or less) for most people.
*[note: 56 hours of GPU training to get the face swap decent. BuzzFeed’s Jared Sosa reportedly built it with FakeApp (then-popular face-swap tool) and Adobe After Effects, using real Obama speech footage and Peele performing the lines (including the lips/mouth performance) before doing comp/cleanup; and audio was trickier back then: if you don’t have Jordan Peele & hire a SAG-scale voice actor for $250 per session, you’re easily at roughly $300 in direct inputs (≈$250 talent ≈$50 compute)]