Higgsfield AI Uses GPT-6 Astra to Extend AI Video From 1 Minute to Over 10 Minutes
Higgsfield AI says its users can now create videos over 10 minutes long, up from a ceiling of around one minute before it adopted GPT-6 Astra. The claim comes from CEO and co-founder Alex Mashrabov in an OpenAI customer story, and it matters because video length and cross-shot consistency have been among the hardest constraints in AI video generation.
What Higgsfield AI announced
Higgsfield AI is a company whose stated mission is to make professional video and image creation accessible to anyone with an idea. According to Mashrabov, the shift from roughly one-minute clips to videos over 10 minutes is the most visible change brought by GPT-6 Astra.
Mashrabov frames the change less as a feature upgrade than as a shift in who can do the work: smaller creative teams can now take on more ambitious projects. In his words:
"What's most important for us at Higgsfield is to see that smaller creative teams can actually be more successful now with the power of models like GPT-6 Astra and GPT-Image-2.5."
Longer videos and the consistency problem
Higgsfield AI's creative agents handle repetitive production tasks, such as generating multiple variations of a shot. With GPT-6 Astra, those agents can also call external tools to handle complex, manual work in 3D modeling.
According to Mashrabov, that tool-calling capability addresses one of creative AI's hardest problems: maintaining consistency across longer videos. Consistency across many shots is what separates a usable long-form video from a collection of disconnected clips, and it is the constraint that historically kept AI-generated video short.
Instructional video generation on Higgsfield AI's YouTube channel
Higgsfield AI runs a YouTube channel that Mashrabov says has almost half a million subscribers, where the company teaches creatives how to use AI models and workflows. Producing that content used to require considerable manual effort.
With GPT-6 Astra, the team has built a workflow that uses AI to generate instructional videos over 10 minutes long. The channel therefore functions as both a marketing asset and an internal demonstration of the longer-video capability.
Education use case in Kazakhstan
Mashrabov says longer instructional video matters to him personally. He supports rural schools in Kazakhstan, and describes an early effort that used GPT-6 Astra's visual understanding to identify topics students may be struggling with, then create videos tailored to what those students need to learn.
This is the only education application described in the source, and no results, metrics, or scale figures for that effort are given.
GPT-Image-2.5 for readable text and infographics
For small businesses producing promotional content, Mashrabov argues that readable text and precise infographics can matter as much as the image itself. He says GPT-Image-2.5 makes that kind of content possible for businesses without large production budgets.
"GPT-Image-2.5 is state of the art and the first model of its kind that can reliably render the text and infographics on top of the imagery."
Text rendering inside generated images has been a persistent weakness of image models, which is why Mashrabov singles it out as the differentiator for commercial promotional work.
Codex and the path from idea to shipped product
Higgsfield AI builds its product through close collaboration between around 200 creatives and 100 engineers, according to the source. Codex gives the creative team what Mashrabov describes as "the power of a really strong software engineer."
Creatives use Codex to turn ideas into software prototypes. Previously, moving those prototypes into production required heavy involvement from front-end and back-end engineers. Mashrabov says GPT-6 Astra now reduces the work needed to bring those ideas into the product:
"GPT-6 Astra can ship production-ready code, which was never possible before."
With less friction between prototyping and production, Mashrabov says Higgsfield AI can develop and ship better products faster.
Voice-to-video with GPT-Live-1
The next target for Higgsfield AI is what Mashrabov calls "voice to movie": a creator describes an idea and watches it take shape as they speak, with AI understanding spoken intention and rendering the vision in real time.
He points to OpenAI's advances in planning, tool calling, image generation, and voice as bringing that experience closer, and describes its voice model as "best in class for voice understanding and just for regular communication." The stated ambition is to make creating a film feel more like describing an idea, with AI handling more of the work in between.
What the source does not specify
The post is a vendor customer story rather than a technical paper or benchmark report. It does not include measured quality metrics for the longer videos, pricing or availability details for any of the models named, the exact resolution or frame rate of the 10-minute outputs, or a release date for the voice-to-video capability. The figures it does provide are Higgsfield AI's own: roughly one minute as the previous video length ceiling, over 10 minutes as the new one, almost half a million YouTube subscribers, and a team of around 200 creatives and 100 engineers.