invideo says that with GPT‑6 Astra, it can plan complex video edits more precisely and apply them at the correct points on the timeline. The tool, which keeps the human editor in control, translates instructions into steps, selects the appropriate tools, executes the process, and verifies the result. According to Sanket, Astra completes complex tasks with fewer reasoning steps and output tokens than previously tested models; it can also retain the original objective as a task becomes longer and more difficult. While the success rate in color correction and grading has increased approximately threefold, the model can choose among approaches such as correction, grading, regeneration, LUT, and isolation. For example, to change the background while preserving a person’s skin tone, it isolates and tracks the person across frames. Astra also codes custom effects from images and descriptions and places them on the timeline; Invideo editors created approximately 50 effects in one day.
Why it matters
This development could affect not only the speed of operations within video editing teams, but also how complex tasks are managed. In particular, the improvement in processes requiring expertise, such as color correction and grading, could allow editors and colorists using these tools to handle more work within the same workflow. Generating custom effects from images and descriptions and adding them to the timeline also brings effect design closer to the editing process, rather than treating it as a separate stage of production. However, keeping decisions and final control in human hands leaves open the question of how closely the operations proposed by the model align with editorial preferences. The real test will be whether these gains can be maintained at the same level across different projects and a wider range of editing needs.
Background
Astra is not a new name in the FikirPilot archive: in the last 90 days, we have published 9 news reports mentioning it; the latest is dated October 3, 2026.