Google Research unveils AI co-director framework for coherent videos up to 10 minutes

Google Research unveils AI co-director framework for coherent videos up to 10 minutes

N
News Editor
2026-09-28 02:51:34
Google Research has introduced an agent framework called AI Co-Director, aimed at addressing identity drift and cascading errors in long-form video generation. The suite includes four components — Co-Director, CANVAS, A²RD, and VQQA — that can stitch short clips into multi-minute story videos while also supporting voice and soundtrack generation. According to the information cited from MarkTechPost, Co-Director scored an average of 81.4 on the GenAD-Bench benchmark and received a human rating of 3.96 out of 5. CANVAS improved background continuity, character consistency, and prop consistency by 21.6%, 9.6%, and 7.6%, respectively. A²RD improved consistency by as much as 30% across videos ranging from 1 to 10 minutes, while boosting narrative coherence by 20%. The system is built on top of Gemini and Veo, but Google Research said the design is model-agnostic and can work with other generators. Outputs also inherit the base model’s SynthID watermarking technology.

Google Research has introduced an agent framework called AI Co-Director, built to tackle identity drift and cascading errors in long-form video generation.

The suite includes four frameworks — Co-Director, CANVAS, A²RD, and VQQA. Together, they can stitch short clips into coherent story videos lasting several minutes, with support for voice and soundtrack generation.

Benchmark results

According to MarkTechPost, Co-Director posted an average score of 81.4 on the GenAD-Bench benchmark and received a human rating of 3.96 out of 5.

CANVAS improved background continuity, character consistency, and prop consistency by 21.6%, 9.6%, and 7.6%, respectively. A²RD raised consistency by as much as 30% for videos ranging from 1 to 10 minutes and improved narrative coherence by 20%.

Underlying models

The system is built on Gemini and Veo, though it is designed to be model-agnostic and can also drive other generators. Its outputs inherit the base model’s SynthID watermarking technology.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
100

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.