Photo by Ivan Samkov on Pexels

Google’s Veo 3, its advanced AI video generator, is reportedly struggling with an unexpected issue: persistent and often nonsensical subtitles. Despite user requests to exclude them, Veo 3 frequently inserts unwanted captions into generated videos, causing frustration and driving up costs for creators.

Users leverage Veo 3 for a range of applications, including creating hyperrealistic short clips for advertisements, ASMR content, and even conceptual film trailers. Noted director Darren Aronofsky recently used the tool to create a short film. The unremovable subtitles are creating difficulties in the creative process.

Frustrated users are resorting to costly workarounds such as regenerating clips (each costing a minimum of 20 AI credits, with credits costing $25 per 2,500), employing external editing software, or cropping videos to eliminate the unwanted subtitles.

While Google has acknowledged the problem and deployed some fixes, reports of the issue persist. Experts at MIT Technology Review suggest the root cause lies within Veo 3’s training data, which likely contains a significant number of videos with embedded subtitles from platforms like YouTube and TikTok. Correcting this would involve substantial retraining efforts. Google states that they are continuously working to improve video creation, especially with text, speech that sounds natural, and audio that syncs perfectly.