Creator Looks Back at Veo-2's Early Limits

A video creator's social media post reflects on how far AI video generation has come since Google's Veo-2, which lacked generated audio and reference tools that are now standard.

A recent post from creator Theo Media revisits Google's Veo-2 model, noting that while its output looks noticeably dated by today's standards, it was considered state of the art when it launched. The post serves as a reminder of how quickly the underlying technology has moved, even over a relatively short span of time. According to the post, Veo-2 shipped without generated audio, meaning any sound had to be added separately rather than produced by the model itself. It also lacked what the creator refers to as OmniRef, a reference-based feature that has since become part of how creators guide newer video models toward specific characters, styles, or scenes. The post also flags character consistency as one of the biggest pain points of that era. Keeping a character looking the same across multiple shots or generations was described as a major challenge at the time, a problem that has since become a key benchmark newer video models are judged against. The original post trails off before detailing what the creator considers the most striking part of the comparison, but the broader point lands clearly: tools that felt cutting edge only a short while ago can quickly come to look primitive as AI video models add features like native audio, reference conditioning, and more reliable character continuity.