OpenAI's GPT-6 Astra enables Higgsfield to generate 10-minute AI videos, up from 1 minute

On October 14, OpenAI announced a partnership with AI video generation platform Higgsfield. After integrating OpenAI's GPT-6 Astra, Higgsfield raised the maximum duration for AI-generated videos from 1 minute to 10 minutes.

According to OpenAI, GPT-6 Astra lets creative agents call external tools and handle complex tasks such as 3D modeling. Higgsfield said this capability helps maintain consistency in characters, scenes, and visual details across long videos, a key challenge in long-form video generation. The workflow has already been used on Higgsfield's YouTube channel, which has nearly 500,000 subscribers; the team has used GPT-6 Astra to produce instructional videos longer than 10 minutes while reducing manual production work.

For images, Higgsfield integrated GPT-Image-2.5, which the company says can reliably render readable text and infographics on images, helping small businesses create promotional materials at lower production costs. Higgsfield is also using OpenAI's Codex to speed up product development. With around 200 creative staff and 100 engineers, the company's creative team can use Codex to turn ideas into software prototypes with less involvement from front-end and back-end engineers.

Higgsfield also disclosed its next-stage direction: exploring "voice-to-video" creation based on GPT-Live-1, where users describe ideas verbally and AI understands the intent to generate video content in real time.