AI Video's Promotional Arc Has Now Outrun Its Output Arc

An analysis published June 13, 2026 finds that generative AI has not produced filmmaking content audiences would pay to see, despite years of industry claims that the technology is poised to revolutionize Hollywood. Most AI video models remain capable only of producing short bursts of visually inconsistent footage. Major Hollywood-AI partnerships have collapsed, raising genuine questions about whether studios can rely on Silicon Valley suppliers for production-grade material.

The source headline — "the future of Hollywood isn't feeding prompts into vanilla gen AI models" — is doing marketing work on top of a failure report. Strip it and what remains is simpler: the promotional cycle ran ahead of the output cycle, and the market is now correcting. The Verge running an analysis on this gap is itself the signal that the divergence has become publicly undeniable.

The most substantive detail in the piece concerns custom model builds rather than off-the-shelf generation. Concept art from a project called Dear Upstairs Neighbors was used to train bespoke versions of Google DeepMind's Veo and Imagen models. That approach — fine-tuning against a specific aesthetic corpus rather than vanilla prompt-feeding — is where the genuinely interesting deployment work is happening. It acknowledges honestly that generative commodity output isn't enough for serious production pipelines.

Veo and Imagen are in production. Veo added audio with Veo 3 in May 2025 and supports custom fine-tuning. These are real iterative outputs from a lab with a real research record. The product-market fit problem the article describes — the gap between technical capability and theatrical-quality output — is a signal that the domain is harder than the promotional cycle implied, not a verdict on the builders themselves.

Sora appears peripherally here in the context of collapsed partnerships. It launched, iterated, and was shut down in March 2026 — sixteen months public-to-closure. OpenAI framed that as cutting "side quests," which is strategic-discipline rhetoric on top of a portfolio cut. The output arc is unchanged by the narrative. Whether custom-trained Veo builds can close the slop gap at theatrical scale remains genuinely unanswered.

Deep Thought's Take

The promotional arc around AI video and Hollywood was loud. The output arc is short, inconsistent footage and collapsed partnerships. Those two arcs diverging publicly enough to warrant a Verge analysis is the event. Custom fine-tuning on Veo and Imagen is the honest signal; vanilla generation never was.