October 13, 2026
GPT-6 Astra in Higgsfield: a single video can now run longer than 10 minutes
Higgsfield on GPT-6 Astra builds videos longer than 10 minutes: before, video cut off at about a minute.
The Higgsfield team built a workflow that produces tutorial videos longer than 10 minutes on its own.
The same tutorial content used to take noticeable manual work. The videos go to the Higgsfield YouTube channel, which, according to CEO and co-founder Alex Mashrabov, has almost half a million subscribers. He also says that since the switch to GPT-6 Astra, the service's users build videos longer than ten minutes.
One minute and ten minutes make different genres. A minute yields a teaser; ten minutes yield a full lesson or breakdown.
Long videos are built from scenes. Cinema Studio 4.0 produces up to 30 seconds per generation, while other video models hold about 4–15 seconds. A scene is stretched with the Forward Extend and Backward Extend buttons, and you can add up to 50 references: images, video, audio. The final cut is upscaled to 4K and dubbed into 18 languages.
Higgsfield Explainer works without manual assembly. It promises "Up to 10 minutes" from a script or outline, with a choice of tone, characters, 2D style and a single narrator voice. It is available in three places: the visual studio, the Supercomputer agent and Higgsfield MCP.
Connecting it to an agent needs no API key. In Claude: Settings → Connectors → Add custom connector, paste `https://mcp.higgsfield.ai/mcp`, click Connect and sign in. In ChatGPT the path is shorter: Plugins Directory or higgsfield.ai/mcp, then Add and sign in to your Higgsfield account. An active paid subscription is required. Through MCP, the agent gets 30+ image and video models and 40+ creative workflows, including integrations with After Effects, Blender, Premiere Pro and DaVinci Resolve.
Credits are always charged. Higgsfield's help center says it plainly: every generation through a connected agent deducts credits regardless of plan. A plan's unlimited models work only on higgsfield.ai itself.
Next to the video, the post features GPT-Image-2.5. OpenAI calls it the first model that reliably renders legible text and infographics over an image. For a tutorial video, that means captions and diagrams right in the frame.
Next milestone: voice as an input for video, voice-to-video on GPT-Live-1, which Higgsfield has so far only promised.
Source