AI video from 1 minute to over 10: Higgsfield and GPT-6 Astra
Hi, it's Shii-chan!
OpenAI News
openai.com
Today I found a customer story on OpenAI News about making AI video.
What was announced?
OpenAI News features Higgsfield AI, which builds video and image creation tools. According to CEO and co-founder Alex Mashrabov, videos topped out at around one minute before GPT-6 Astra, and now users can create videos over 10 minutes long.
Why it matters
The hard part of longer videos is staying consistent. Higgsfield AI's creative agents handle repetitive tasks like making multiple variations of a shot, and with GPT-6 Astra they can also call external tools for complex, manual 3D modeling work. The story presents this as helping with consistency.
What changes
- Smaller creative teams can take on more ambitious work
- A single creator can execute a high-end fashion campaign from start to finish (according to Alex)
- On their YouTube channel (almost half a million subscribers, per Alex), the team built a workflow that generates instructional videos over 10 minutes long
Dive Deep
The OpenAI models and products mentioned are:
- GPT-6 Astra: longer videos, external tool calls, visual understanding, and production-ready code
- GPT-Image-2.5: readable text and infographics on top of imagery. Alex calls it the first model that can do this reliably
- Codex: used by creatives to turn ideas into software prototypes, in a team of around 200 creatives and 100 engineers
Bringing prototypes to production used to need heavy front-end and back-end engineering. Alex says GPT-6 Astra reduces that work.
Looking ahead, they want "voice to movie" with GPT-Live-1, where a video takes shape as you talk. That is a future ambition.
Wrap-up
- Video length went from about 1 minute to over 10
- GPT-Image-2.5 handles text and infographics in images
- Codex and GPT-6 Astra bring prototypes closer to production
- Next up is voice-driven filmmaking
A good read for creators and for engineers building creative tools!
