Pika launches Generative UI experiment with voice-controlled HTML layouts

• Pika opened a new experimental track with pika_labs' Generative UI announcement, describing it as the first release in a broader "Pika Experiments" series.
• The experiment is a voice-controlled interface that listens, reads context, and generates a fresh HTML layout for each response, according to pika_labs' product description.
• Pika is framing the release cautiously, with pika_labs' follow-up note saying these experiments are intentionally rough around the edges.
• Access is already live through pika_labs' GitHub post, which points users to a public repo and says the setup requires Pika MCP.

You can watch Pika's demo post turn spoken input into different on-screen compositions, open the GitHub repo for the experiment itself, and see in Pika's caveat that the company wants this read as early work, not a polished product ship.

Pika says Generative UI skips fixed templates. The agent listens to voice input, analyzes the current context, and picks the visual composition that fits the response.

That puts the interesting part in the layout layer, not just the model response. In pika_labs' announcement, the system is described as generating HTML dynamically each time, which makes the interface itself part of the output.

Pika has already published the experiment in its GitHub repo, and pika_labs' access note says the path in runs through Pika MCP.

The company is also signaling how to read this release. pika_labs' disclaimer calls these experiments intentionally rough around the edges, which makes this look more like a public lab notebook than a finished feature rollout.