Pika Post-Trains on MiniMax H3 and Open-Sources the Pika H3 Flex Video Generation Model
@pika_labs post-trained on Minimax’s H3 and released Pika H3 Flex, reducing the number of video-generation sampling steps from 50 to 6. Generation time is much shorter, and quality has even improved. Very cool work, and it’s open source too.
AI BriefingPika Labs has post-trained MiniMax H3 to create and open-source Pika H3 Flex, reducing the number of sampling steps from 50 to 6, greatly shortening generation time while improving quality.
Source: Kazike · x.com