MiniMax’s H3 video generation model, whose weights were released in August this year, has been adopted by a joint Nvidia-MIT team, Pika and fal, spawning derivative products for video enhancement, inference acceleration and professional content creation. In just over three weeks since release, H3 has been downloaded more than 24 million times and has generated over 300 derivative models, ranking first among open-weight video models in third-party evaluations. The joint Nvidia-MIT team launched a 2K/4K high-definition video enhancement solution based on H3 called Open-H3-Regenerate-2K, which cuts the generation time for a 5-second 2K video on a single B200 from 87 seconds to 22 seconds; the team had previously achieved a 3.95x inference speedup on an 8-GPU GB200 cluster within 4.5 hours of H3’s release and later adapted the model to hardware including the RTX 5090 and DGX Spark, while launching Sol-H3 and Sol-H3-VAE. Pika produced two lightweight adapters of about 2.8GB each that make up Pika H3 Flex, enabling video generation with fewer inference steps and enhanced control for professional creative scenarios. fal’s H3 Max, built on H3, topped multiple leaderboards after launch. H3 uses a tiered community license: free commercial use for users with annual revenue below $20 million, while