Pika drops audio-driven video model with hyper-real expressions

Pika just unveiled a new audio-driven performance model designed to deliver hyper-realistic facial expressions in near real-time.

Six-second HD video at 20x lower cost

According to Pika, the new model can generate HD video of any length and style in six seconds or less.

At the same time, the team claims to have made the generation process 20 times faster and 20 times cheaper than previous approaches.

The feature is scheduled to roll out soon inside the Pika social app. Pika noted that the goal is to make AI video accessible and fun for everyday users to connect with each other, rather than serving solely as a creation tool.

Why it matters

Audio-driven avatar and lip-sync tools have long been plagued by robotic expressions and painful rendering wait times. If Pika can reliably deliver expressive, full HD video in under six seconds, it transforms AI video from a deliberate editing project into an instant messaging format.

From an infrastructure perspective, cutting compute costs by 20x is essential if any company hopes to power video generation inside a consumer social app without burning through endless server budgets.

That said, claiming that videos of any length can be rendered within six seconds sounds almost too good to be true. We think the real test will be whether the output maintains high consistency and realistic motion once everyday users get their hands on it inside the app.