The AI Video Arms Race Got Loud in 2026

AI video spent 2026 moving from demo to deliverable, and the shakeout was brutal. The biggest story wasn't a launch; it was a shutdown: OpenAI folded the Sora app in April 2026, citing compute costs and weak engagement, and a reported Disney partnership collapsed with it.

The survivors got serious about the problem everyone hit: consistency. Runway's Gen-4.5, launched early in the year, added a World Consistency system to keep characters steady across shots, and led independent benchmarks on that metric. Google's Veo 3.1, out in January, lets you anchor a subject with up to three reference images. Both labs shipped near-identical fixes within months, a clear sign character drift is the real blocker for production video.

Studios are placing bets. Lionsgate took an equity stake in Runway in June 2026 to build AI short-form series from its IP, and Google DeepMind partnered with A24 the same month. The deals tie the model provider to the output, a different accountability than a software license.

Meanwhile ElevenLabs pushed Eleven v3, a voice model with emotional nuance that makes AI narration less robotic, and Luma shipped Dream Machine 3.5 with keyframe control. The stack is filling in: voice, motion, consistency, all maturing at once.

The takeaway for creators: the tools can now produce watchable footage, but the governance, who reviews the 70,000 clips it takes to make one ad, is where the real cost lives. The Coca-Cola 2025 case, where positive sentiment dropped after an AI campaign, is the cautionary tale everyone cites and few heed.