The open-source MiniMax H3 model has surged in popularity following rapid speed enhancements, reshaping the economic and technical landscape of video generation with faster, more affordable outputs.
MiniMax H3 has rapidly become one of the most closely watched video generation models after its open-weights release in late July, with developers and users now pushing the system in several directions at once. According to the model’s own site and MiniMax’s announcement, H3 is designed as a general-purpose multimodal system that can work across text, images, video and audio, while producing clips with native stereo sound at up to 2K resolution and lengths of up to 15 seconds.
The latest wave of attention has come not only from the base model itself, but from community modifications that prioritise speed. ASCII.jp reported that ComfyUI downloads of MiniMax H3 have approached 20 million, a sign of how quickly the model has spread through creator workflows. The same report highlighted fal’s MiniMax H3 Max, a customised version that can generate a 15-second clip in about 13 seconds while preserving much of the original output quality.
That speed matters because it changes the economics of practical use. fal says its 768p service is priced at $0.08 per second, or $1.20 for a 15-second clip, with a temporary half-price promotion running until 1 September. ASCII.jp compared that with Seedance 2.0’s audio-enabled standard pricing of $0.30 per second, or $4.55 for 15 seconds, underlining why users are taking notice of the new H3 variants.
The technical pitch behind H3 Max is that the model has been post-trained specifically for the faster configuration, rather than simply being forced into a shorter inference path that would damage quality. fal has said the system is optimised for 768p output and fewer inference steps, with design choices shaped around its own serving engine. Alongside H3 Max, other acceleration methods such as FastH3 are also emerging, suggesting that the real story is no longer just model quality, but how far video AI can be sped up without falling apart.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
