Enhancing Agentic Reasoning and Planning

Invideo utilizes GPT-6 Astra as an agentic engine to manage the technical and creative demands of video editing. The model excels at translating high-level editor instructions into precise, frame-level sequences. A key performance gain noted by CEO Sanket Shah is the model's efficiency in reasoning; Astra completes complex tasks using fewer reasoning steps and output tokens compared to previous models, allowing it to maintain the editor's original intent over longer, more complex workflows.

Improving Color Grading and Correction

Color grading often requires navigating overlapping techniques like isolation, LUT application, and regeneration. Previously, Invideo faced high failure rates when automating these tasks, particularly when complex requirements—such as changing a background while preserving specific skin tones—were involved. By integrating GPT-6 Astra, the company achieved a 3x increase in the success rate for color-grading and correction tasks, demonstrating the model's superior ability to handle multi-step visual processing.

Rapid Custom Effect Generation

Beyond standard editing, Astra enables the creation of custom effects based on natural language descriptions and visual references. The agent generates the necessary code for these effects, places them on the timeline, and exposes controls for human editors to refine the output. This capability significantly accelerates production, with Invideo editors successfully creating 50 custom effects in a single day. This workflow preserves the human editor's creative control while offloading the tedious, frame-by-frame execution to the AI.