Pika H3 Flex opens six-step video-and-audio generation for local use
Pika released H3 Flex, a post-trained MiniMax H3 model family that generates video with synchronized audio in six sampling steps. Public weights, code and documented paths cover text-to-video, first- and last-frame generation, and reference-conditioned video at native 480p or 768p with up to 12 image, video or audio references.
Why it matters: The release gives video teams an inspectable local path for rapid multimodal iteration instead of limiting the workflow to a hosted creator. The useful boundary is narrower than the launch headline: local artifacts are adapters derived from MiniMax H3, while…