Higgsfield takes AI video from 1 minute to over 10

Higgsfield has raised its AI video length cap from roughly one minute to more than ten, a change that meaningfully expands what creators can produce in a single generation session. Until now, the one-minute ceiling forced users to stitch together multiple clips manually - a workflow that often introduced visual inconsistencies and interrupted narrative flow. Longer native generation reduces that friction considerably.

The update leans on GPT-6 Astra for the extended video backbone, with the model handling coherence across a much greater number of frames than its predecessor could manage. Maintaining consistent characters, lighting, and scene continuity over ten-plus minutes is a harder problem than doing so over sixty seconds, and Higgsfield is citing GPT-6 Astra as the component that makes it tractable at this length.

On the visual fidelity side, GPT-Image-2.5 is now integrated to give creators finer control over how specific elements look within a generation. This kind of precise image grounding matters when a creator needs a character or object to remain recognizable across a long runtime - something that has been a persistent weak point in AI video tools broadly. The inclusion of a dedicated image model suggests Higgsfield is treating visual consistency as a first-class concern rather than an afterthought.

Perhaps the most forward-looking addition is GPT-Live, which opens a path toward voice-driven filmmaking - letting creators direct or narrate their way through a production rather than relying solely on text prompts or manual parameter adjustments. The practical implications of that workflow are still emerging, but it points toward a future where the gap between having an idea and having a finished video continues to shrink. Taken together, the three updates reflect a broader push in the AI video space to move beyond short clips and toward tools that can support more complete creative projects.