Grok’s seven-minute AI ‘Odyssey’ shows convincing shots and very little story

Three months after Elon Musk promised a full-length, “historically accurate” Grok version of The Odyssey, what has appeared is seven minutes, now billed as Episode 1 of a series. It was made with Grok Imagine, the video and image generator from Musk’s AI company, with Wonder Studios and director Dev Walker.

The timing invites comparison. Christopher Nolan’s Odyssey, a largely practical production shot in six countries with a cast and crew of thousands, opened in July, sold out Imax screenings for months and is seen as a likely Academy Awards front-runner. Musk had attacked that film over its casting. Nolan has called himself a techno-skeptic and said he had never seen a technology so rejected by the public as AI. So this clip is framed as an answer to Hollywood, which raises the stakes for how AI filmmaking is judged.

TechCrunch’s reviewer, who expected much worse, said some shots could pass for real actors on a real sea. A year or two ago, AI video at this length would have fallen apart. For independent creators, previsualization and small studios without a six-country budget, that capability is genuinely useful. Grok’s underlying models have been improving on price and performance (Grok 4.7).

The reviewer’s real verdict is that the footage feels airless: endless voiceover, a parade of bearded faces, stilted lines, no momentum and no reason to care. He said he would have skipped it as a video game cutscene. That is the persistent gap in generative video. Models have become good at pixels and still have little to say about people.

This is what happens when a tool demo is asked to carry a culture-war argument. The project began as a rebuke to a film over its casting and arrived as something plausible frame by frame and hollow overall. Craft didn’t lose to the machine; the machine was pointed at a grievance instead of a story.

When anyone with a prompt can produce an epic that looks like a studio picture, the scarce thing is no longer footage. It is taste, writing, performance and trust. Platforms risk being flooded with cheap, glossy, empty content that crowds out human work, and “historically accurate” could become a label for whichever version of history the prompter prefers, the same question that hangs over Grokipedia’s “fact-checked by Grok” stamps (coverage).

The best case for AI video is as an instrument in filmmakers’ hands. This episode reads more like the instrument playing itself. Whether later episodes add real creative input will show which path the project takes. More in our generative AI coverage and xAI coverage.