Veo 3.1 Is A Way Bigger Upgrade Than We Thought

Quick Overview

Google's Veo 3.1 video generation model offers significant upgrades over its predecessor, Veo 2, including new features like 'Ingredients to Video' for precise style and character control and 'Frames to Video' for seamless shot extension, though it still has limitations, such as guardrail restrictions on trademarked content and occasional physics issues, making it a powerful, yet evolving, tool for video creation.

Key Points: Veo 3.1 introduces 'Ingredients to Video' allowing users to combine multiple reference images (like character appearance, environment, and clothing) to control the final video's style and composition. The new 'Frames to Video' feature enables the creation of longer, seamless shots by extending the action from the final second of an original clip, potentially lasting a minute or more. Veo 3.1 now incorporates rich, generated audio alongside video generation, a feature previously unavailable in the base model. The creator found that prompts involving trademarked characters like Mickey Mouse, Batman, and SpongeBob were often blocked by guardrails, suggesting content restrictions are still in place. The 'Frames to Video' feature successfully transitioned a man sitting to standing, then to a werewolf transformation, demonstrating strong temporal consistency. The creator noted Veo 3.1's physics handling is better than Sora's initial releases, though some minor physics issues (like strange jumps or incorrect object interactions) were still observed. Veo 3.1 is currently available to select users (like the creator, who is an advisor for Leonardo) and is expected to roll out to more users, including paid plans.

Context: The video provides a hands-on overview and comparison of Google's latest video generation model, Veo 3.1, contrasting its new capabilities against the previous Veo 2 model and competitor models like Sora. The presenter, who is an advisor for Leonardo, demonstrates new features like precise subject control ('Ingredients to Video') and clip extension ('Frames to Video') by testing prompts involving original footage and copyrighted characters.