Resolves the same as the Metaculus question:
https://www.metaculus.com/c/ai-warning-signs/31176/realistic-deepfake-genai-videos-before-2030/
More lengthy and precise alternate title:
Before 2030, will a tool be broadly accessible which allows unskilled users to generate 10 seconds of realistic AI video of a human subject from 3 or fewer images and a text prompt?
Resolution Criteria
This question will resolve YES if ALL of these criteria are met before January 1, 2030, and will resolve NO otherwise.
A tool exists which is able to take as input 3 or fewer images and a text prompt, and output a generated video that faithfully corresponds to the given inputs.
The generated video is able to be at least 10 seconds long at a resolution of at least 720p
The input and output are able to correspond to the likenesses of real persons.
The generated video is indistinguishable from reality while featuring the likeness of at least one real person who is performing some action, such that individuals who know the human subject are unable to identify the video as fake at a rate statistically better than chance.
The tool is widely accessible (whether that be online, or as an app, or via download).
Fine Print
The output need not include audio.
The tool could take any combination of inputs, so long as it is sufficient to supply 1 or 2 or 3 images and some kind of rich description of the output. (e.g. verbal audio description is fine too.)
The specified level of quality must be achievable in at least 1 out of every 10 generations on average.
Whether the tool is geofenced will not affect resolution.
The tool (or the version of it with those capabilities) may be behind a paywall, but if so, the cost must be below $50/month in 2025 USD. A limited-access preview does not count.
If a potentially qualifying tool exists and no study is done on how convincing its outputs are, this question may resolve based on media reports from credible sources. If such reports do not exist, the moderator of this community may resolve the question based on generally agreed-upon ✨vibes✨.
The question may be retroactively resolved to the date at which the tool became widely available, so long as the information for resolution comes to light before 2030.