AIbaseUpdated

10-Minute AI Long Videos Are Here: GPT-6 Astra Takes Over Higgsfield, So Characters and Scenes No Longer "Cross Over"

OpenAI has partnered with Higgsfield to raise the maximum length of AI-generated video from 1 minute to 10 minutes. The key enabler is Higgsfield's integration of GPT-6 Astra, which lets "creative agents" call external…

The latest collaboration between OpenAI and the AI video generation platform Higgsfield has pushed a previously rigid boundary a big step forward—the maximum length of AI-generated video has jumped straight from 1 minute to 10 minutes. Underpinning this leap is the GPT-6 Astra that Higgsfield has integrated. Its most crucial capability is enabling "creative agents" to call external tools to handle complex work, which amounts to giving video generation a little director who can orchestrate things on its own, thereby ensuring that a long video stays consistent throughout in character appearance, scenes, and visual details—no more "cross-over" accidents where a person looks like one thing early on and has a different face later.

Higgsfield has already put this workflow into real-world practice. Its YouTube channel, which has nearly 500,000 subscribers, has begun producing educational videos longer than 10 minutes; work that used to require people to painstakingly stitch together and proofread footage has been swallowed up by the model in one go, saving considerable manpower. Meanwhile, Higgsfield has also integrated GPT-Image-2.5, which is dedicated to reliably rendering readable text and infographics—especially friendly to small businesses, meaning they no longer have to go to great lengths for promotional materials and can generate decent finished videos at low cost.

An even cleverer step is on the development side. With the help of OpenAI's Codex, Higgsfield has accelerated its product iteration speed: an idea in a creative person's head can now be translated almost directly into a software prototype, and engineers no longer have to get involved from scratch at every stage. In the next phase, Higgsfield plans to explore a "voice-to-video" real-time creation model based on GPT-Live-1—meaning that in the future, you could speak into a microphone and watch images grow in sync. Going from 1 minute to 10 minutes is only the opening act; what AI video really aims to seize is the territory of long-form content production, which has always been held in human hands.

Original source

AIbase

Content notes

Original publication and rights belong to the source.

Machine translation · Refer to the original