Higgsfield takes AI video from 1 minute to over 10

An OpenAI customer story says Higgsfield AI, led by CEO Alex Mashrabov, used GPT-6 Astra to push AI-generated videos from about one minute to more than ten minutes, with GPT-Image-2.5 improving readable text and infographics for small-business creatives. Higgsfield’s agents handle repetitive shot variations and can call external 3D tools to improve consistency across longer pieces.

The company reports nearly half a million YouTube subscribers for instructional content now produced with longer AI workflows, and describes Codex plus GPT-6 Astra shortening the path from creative prototypes to production code across a staff mix of roughly 200 creatives and 100 engineers. Mashrabov points next to “voice to movie” experiences using OpenAI’s voice models such as GPT-Live.

Longer AI video is the obvious next boss fight after pretty five-second clips. Consistency across ten minutes is where most systems still fall apart; Higgsfield’s claim that Astra-class agents plus tool calling get them there is worth watching—on their channel and in customer work, not only in a partner blog.

We’re glad small teams can attempt fashion-campaign-scale work without a huge production department. We’re also clear-eyed about provenance, labor displacement in postproduction, and the risk that “instructional” AI video floods platforms faster than quality signals.

OpenAI’s post is a customer win reel. Treat performance claims as directional. The interesting structural point is creatives using Codex to prototype product while Astra shortens the engineer handoff—blurring studio and software company.

If voice-to-movie lands as more than a demo, the interface for filmmaking changes again. Until then, ten-minute coherent video with controllable brand text would already be enough to rearrange a lot of agency scopes.

AI video tools have been limited by short duration and identity drift. Agent frameworks that call external 3D and editing tools are one path to longer, more consistent outputs.