Higgsfield Stretches AI Video From 1 Minute to Over 10 Minutes
OpenAI has published a case study on how Higgsfield uses its models. The company is applying them to long-form educational videos and promotional images, while voice-driven video creation remains at the concept stage.
On October 13, OpenAI published a case study on how Higgsfield, an AI video production service, is using its models. Higgsfield’s CEO said that with GPT-6 Astra, the length of video that can be made has grown from roughly 1 minute to more than 10 minutes.
How do you build a video longer than 10 minutes?
When producing long-form video, GPT-6 Astra acts as the connective tissue between the many separate tasks in the creative pipeline. An AI agent, meaning a system that handles work on its own, calls external tools and hands off 3D modeling work to them. OpenAI said this approach helps keep visuals consistent even across longer runtimes.
Using this method, Higgsfield has also built a workflow for producing educational videos that run longer than 10 minutes. It is a case of expanding beyond short clips into educational content.
Which model handles promotional images?
On the image side, the focus has been on how accurately promotional materials that small business owners can use are rendered. According to OpenAI, GPT-Image-2.5 is used to render legible text inside images and to produce precise infographics, meaning visuals that organize information into illustrations and charts.
Image: OpenAI
Codex comes in when the creative team’s ideas need to be turned into working software. OpenAI explained that the creative team builds prototypes with Codex, and GPT-6 Astra reduces the work required to carry those prototypes into the actual product. In other words, the company’s use of AI has widened from producing finished videos and images to developing the production tools themselves.
Roughly 200 creators and 100 engineers work on Higgsfield’s product development. The CEO said the company’s YouTube channel has about 500,000 subscribers.
When will voice-driven video arrive?
The idea of describing a concept out loud and having AI understand it and apply it to the video in real time is still at the conceptual stage. Higgsfield said it is aiming for voice-based video production built on GPT-Live-1. OpenAI did not say when the feature would launch or become available.
Image: OpenAI
The range of content being made with AI has stretched from short clips to long-form educational material and business promotional imagery. For educational content creators, video length is the number to watch; for small business owners, it is how well text and information are rendered inside images.