AI news · October 14, 2026

Higgsfield says its AI videos now run past 10 minutes

videocreative-toolsagentsproduction

Higgsfield reports video runs longer than 10 minutes, up from about one minute before GPT-6 Astra; creators can try a short storyboard and judge continuity before changing tools.

reported video length
10+ min
reported creatives
200
reported YouTube subscribers
500K
Made With Models illustration for this story

OpenAI's October 13 customer story describes how Higgsfield moved from short generative clips to videos longer than 10 minutes. The company says creative agents can create shot variations, call outside 3D tools and keep work moving across a larger production. Higgsfield also uses GPT-Image-2.5 for text and infographics. The company reports about 200 creatives, 100 engineers and a YouTube channel approaching 500,000 subscribers. These are company-reported operating figures, not an independent production study.

The useful shift is from a single clip to a repeatable sequence of shots. Long output is valuable only if characters, camera intent, sound and graphics stay usable across edits. A creator can test that today with a short storyboard, three shot changes and one revision request. Measure the number of usable shots, correction time, export cost and final review time against the current process. If the tool creates more fixing work than it removes, longer output is only a bigger file.

What you can do with it

Open Higgsfield and start with a 30 to 60 second storyboard that has three shots and one visual change. Ask for two variations, make one revision and record continuity errors, usable minutes, render time and total cost. Keep a human check for claims, likeness and rights before any public use.

Our take

The story is useful because it describes a production workflow rather than only a model score. The reported ten-minute result is a vendor case study, so it does not guarantee the same result for every creator. The commercial test is simple: does the tool produce accepted scenes faster than a shorter clip workflow plus manual editing?

Source: OpenAI / Higgsfield — Made With Models writes the brief; the reporting is theirs.