On September 23, 2026, invideo reported that integrating GPT‑6 Astra into its agentic video editor improved its ability to plan and execute complex edits. The company said the model helps place changes precisely on the timeline and reduces the manual planning burden for editors.
Frame‑level planning and reasoning
Sanket Shah, CEO of invideo, described GPT‑6 Astra’s planning as “quite stunning,” saying the model can plan edits with frame‑level accuracy. According to Shah, Astra completes complex operations using far fewer reasoning steps and output tokens than the other models the team tested, and it can follow multiple instructions without losing the original objective as tasks grow longer.
Color grading, correction and custom effects
invideo reported a roughly threefold improvement in success rates for color‑grading and color‑correction tasks after adopting Astra. The company noted that color work often requires choosing between overlapping approaches—correction, grading, regeneration, LUTs, and isolation—and that Astra handles those tradeoffs better. As an example, invideo pointed to background changes that must preserve a person’s skin tone: the agent isolates and tracks the person across frames before applying color changes elsewhere.
The company also said Astra can transform descriptions and visual references into editable custom effects. invideo reported that a small group of editors used the model to create about 50 custom effects in one day. The agent reportedly can code an effect tailored to footage, place it on the timeline, and add controls that let editors refine the result while keeping the final decisions in human hands.
Company details provided with the report list invideo as a startup operating in Asia‑Pacific & Oceania within the Media, Entertainment & Sport industry, and noting use of the API product. The account specifies the date of the report as September 23, 2026 and the 50‑effect and threefold success‑rate figures as those reported by invideo and its CEO.
Original source: OpenAI News