OpenAI says Higgsfield can make instructional videos longer than 10 minutes

OpenAI says Higgsfield has built a GPT-6 Astra-based workflow that generates instructional videos longer than 10 minutes for its YouTube channel. Higgsfield CEO and co-founder Alex Mashrabov said its earlier process topped out at roughly one minute.

According to OpenAI’s customer story, the workflow lets Higgsfield produce longer instructional content while maintaining consistency across the output. The account comes from OpenAI and Higgsfield and includes no independent benchmarks of output quality, production time, cost or required human intervention.

OpenAI says Higgsfield’s creative agents handle repetitive production work, while GPT-6 Astra lets them call external tools for complex manual tasks, including 3D modeling. Mashrabov connects that tool use with consistency in longer videos. The source does not say whether a finished video is a single model generation or an orchestrated sequence of generation, editing and external-tool steps.

A separate deployment of GPT-6 Astra for days-long quantitative research offers context on the model’s use in extended workflows, but does not verify Higgsfield’s video claims.

A separate Higgsfield Academy lesson documents a full 10-minute video workflow and recommends a 10-minute length. Its example uses 100 scenes in one visual style generated from one prompt. The lesson does not identify GPT-6 Astra or establish the earlier roughly one-minute limit.

OpenAI presents voice-to-video as a future direction, not a released Higgsfield capability. Mashrabov describes an ambition for creators eventually to speak an idea and see it rendered in real time. The story cites GPT-Live-1 as part of that path, but gives no availability date and documents no shipped Higgsfield voice-to-video product or demonstrated real-time movie-generation workflow.