Inspired by this:

https://manifold.markets/ScottAlexander/in-2028-will-an-ai-be-able-to-gener#afxuogp7anb

As per Musk's tweet, "really good" means a substantial increase over merely "watchable," but not to the level of high budget studio blockbuster. It must be a full length "movie," not merely a long video (e.g. video essay), or a short cohesive narrative/piece of art along the lines of a student film. And it must be made by Grok, although very modest human help is acceptable, so no need for 1-shot from the prompt.

This is obviously an inherently subjective market, but I will try my best to be fair and impartial. Now that the market is "initialized," I will not trade in it.

Fine print:

A movie
- Minimum length of 1 hour 15 minutes.
- Must be 1 cohesive narrative with continuous characters, plot elements, scene locations, etc.

Of "very good" quality
- Must be good enough for Musk to hype it as "very good" and not be clowned on by most of Twitter or the internet at large.
- Should be extremely impressive to neutral/mildly skeptical towards AI (as of 2025) Manifold users; anything more than rare, minor AI gen artifacts will invalidate.

Made by Grok

- The footage and sound must be 99+% generated by Grok (or other XAI video/audio generation AI); this is really just so that stuff like famous audio clips or similar content could be included such as the Wilhelm Scream.
- It can be made via multiple prompted video generations (minimum length 10m each, max 10 generations in movie) stitched together w/ light editing, but only if this doesn't break visual continuity of characters, scenes, etc. in a way that's obviously AI gen.
- It can't be meticulously micro-prompted & edited by a (team of) human(s) simply using Grok as a video/sound generator; this is to prevent weird scenarios where an insane amount of context is given to Grok cumulatively such that Grok is no longer doing most of the creative work.
- If video/audio can be fed into Grok's prompt(s), this must be less than 1% of the total movie length, except for if previous Grok video generations are fed into subsequent prompts (if this works, it's basically just a form of scaffolding).
- Similarly, lines of dialogue can't be primarily provided by the prompt(s): less than 10% of the movie dialogue can be word for word from the prompt(s).

I'm open to minor clarifications if anything is too vague or controversial, e.g. running a poll on if a submitted video is "very good", with the goal being a non-controversial resolution.