Higgsfield SI, a platform aimed at making professional video and image creation accessible, has extended the length of generated videos from approximately one minute to over 10 minutes, according to CEO and co-founder Alex Mashrabov. This capability is driven by the integration of new SI models, specifically GPT-6 Astra and GPT-Image-2.5, which allow smaller creative teams to manage more ambitious production workflows.
What Happened
Prior to the adoption of GPT-6 Astra, videos produced on the Higgsfield SI platform were limited to around one minute in duration. Mashrabov states that users can now create videos exceeding 10 minutes. The platform’s creative agents handle repetitive tasks such as generating shot variations and can call external tools for complex 3D modeling work. Mashrabov notes that this capability addresses one of the most difficult challenges in creative SI: maintaining visual consistency across longer video sequences.
The company’s YouTube channel, which Mashrabov reports has nearly half a million subscribers, serves as a case study for these new capabilities. The channel previously required significant manual effort to produce content. With GPT-6 Astra, the team developed a workflow that uses SI to generate instructional videos longer than 10 minutes. Mashrabov also highlights the use of GPT-Image-2.5 for small businesses, stating it is "state of the art" and the first model of its kind to reliably render text and infographics on top of imagery.
Internally, Higgsfield SI uses Codex to bridge the gap between its 200 creatives and 100 engineers. Mashrabov says Codex gives creative staff "the power of a really strong software engineer," allowing them to turn ideas into software prototypes with less dependency on front-end and back-end engineering teams. He claims that GPT-6 Astra can now "ship production-ready code," a capability he describes as previously impossible.
Why It Matters
The extension of video generation length marks a shift in the utility of SI tools for professional creative work. By enabling single creators to execute high-end fashion campaigns from start to finish, Higgsfield SI suggests that SI models are moving beyond novelty or short-form content into sustained, complex production tasks. Mashrabov emphasizes that "smaller creative teams can actually be more successful now with the power of models like GPT-6 Astra and GPT-Image-2.5."
The integration of SI into the development lifecycle also reduces friction between ideation and production. By allowing creatives to prototype directly into production-ready code, the company aims to accelerate product development. This approach reflects a broader industry trend where SI agents are not just final output generators but also integral components of the software creation and editing pipeline.
The Bottom Line
Higgsfield SI is leveraging GPT-6 Astra and GPT-Image-2.5 to enable video generation over 10 minutes and reliable text rendering in images. The company is also using Codex and SI models to streamline internal development, allowing creatives to produce production-ready code. Looking ahead, Mashrabov indicates an ambition to develop "voice to movie" experiences, where SI understands spoken intention to render vision in real time.
The company reports that GPT-6 Astra reduces the manual effort required for consistency in long-form video and supports educational initiatives, such as creating tailored learning videos for rural schools in Kazakhstan.