Google´s Veo 3 video generator plagued by persistent subtitles issue

Google´s latest generative video tool, Veo 3, is drawing criticism for automatically adding garbled subtitles, even when users ask for none—a major snag for creatives embracing Artificial Intelligence video tools.

When Google unveiled Veo 3, its latest generative video model, anticipation among video creators soared. The tool—now able to generate not only visuals but also sounds and dialogue—enabled a wave of self-styled film trailers, commercials, and creative shorts. Acclaimed director Darren Aronofsky debuted ´Ancestra,´ a film produced with Veo 3, a testament to its creative capabilities. Google DeepMind chief Demis Hassabis compared this leap in realism to film’s transition out of the silent era.

However, Veo 3’s launch was quickly overshadowed by a persistent technical flaw: the platform often overlays nonsensical or garbled subtitles on generated videos, regardless of whether users request captions. Creators found the only remedies—regenerating expensive clips or scrubbing subtitles with external tools—frustrating and costly. Despite Google’s public assurances of a fix in early June, reports from the product’s Discord community suggest that the problem remains unresolved. With each generation costing users valuable credits, the stakes are high for those aiming to produce professional content, but plagued by outputs rendered unusable by intrusive text.

Experts believe Veo 3’s relentless subtitle insertion likely stems from its training data, which almost certainly includes YouTube and TikTok videos—many of which already feature embedded captions. Because these captions are part of the video frames rather than layered as metadata, filtering them out before training would require a laborious overhaul. Negative prompts like ´no subtitles´ are known to be less effective in generative Artificial Intelligence systems. Google acknowledges continued efforts to refine Veo 3, urging user feedback, but the process may require retraining the entire model with cleaner data—a significant undertaking. For some artists and filmmakers, Veo 3’s subtitle glitch illustrates the penalties of tech companies prioritizing rapid product launches over reliable user experience. For now, Google’s race to dominate artificial video generation is marred by frustrating, unrequested subtitles.

66

Impact Score

Samsung completes hbm4 development, awaits NVIDIA approval

Samsung says it has cleared Production Readiness Approval for its first sixth-generation hbm (hbm4) and has shipped samples to NVIDIA for evaluation. Initial samples have exceeded NVIDIA’s next-gen GPU requirement of 11 Gbps per pin and hbm4 promises roughly 60% higher bandwidth than hbm3e.

NVIDIA and AWS expand full-stack partnership for Artificial Intelligence compute platform

NVIDIA and AWS expanded integration around Artificial Intelligence infrastructure at AWS re:Invent, announcing support for NVIDIA NVLink Fusion with Trainium4, Graviton and the Nitro System. the move aims to unify NVIDIA scale-up interconnect and MGX rack architecture with AWS custom silicon to speed cloud-scale Artificial Intelligence deployments.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.