How invideo uses GPT-6 Astra to speed up color grading

Invideo is an agentic video editor that keeps the human editor in control while an AI agent plans and executes edits. With GPT-6 Astra, the editor’s direction is translated into a sequence of steps, tool choices, execution, and verification. CEO Sanket Shah says Astra can plan a particular edit with frame-level accuracy and uses far fewer reasoning steps and output tokens to complete complex work than the models invideo previously tested. It can also carry out several instructions without losing the original objective as the task grows longer and more complex.

The clearest reported gain is in color work, where the agent chooses from overlapping approaches such as correction, grading, regeneration, LUTs, and isolation. Changing a background while preserving skin tone, for example, requires isolating and tracking the person across frames before applying color elsewhere. Earlier, color-grading and color-correction tasks had very high failure rates; with Astra, the success rate improved about three times. Astra also helps turn descriptions and visual references into custom effects: the agent codes an effect for the footage, places it on the timeline, and adds controls so the editor can refine it. A few Invideo editors used the model to create about 50 effects in one day. The stated benefit is that creative professionals keep control of the story, taste, and final cut without losing hours to planning and frame-by-frame work.

How invideo improves color grading 3x with GPT‑6 Astra

View Original