The Next Frontier of AI Video Is Control
Thursday, 17 September 2026 · 2 min read · Listen to the episode ↗
In this episode, we explore the groundbreaking advancements in AI video generation, particularly focusing on the H3 Max model's impressive speed and quality improvements. The discussion highlights the model's unique capabilities, including real-time video generation and enhanced controllability, which are reshaping creative possibilities. We also delve into the rising demand for open-source solutions and the implications for Hollywood studios as they prepare to significantly increase their AI usage in production workflows.
The H3 Max model marks a pivotal moment in consumer AI video generation, showcasing impressive capabilities in speed and quality. The H3 Max Turbo can generate a five-second video in just 1.5 seconds, indicating a shift towards real-time applications in AI video.
The Minimax H3 model stands out as the first truly open-source, next-generation video model, responding to a rising market demand for video generation. Despite facing compute constraints since April, system-wide optimizations could potentially enhance model speed by 2x to 3x while preserving quality.
The H3 Max model has achieved a remarkable 35x speedup over its predecessor, primarily through post-training adjustments that streamline the number of steps required for video generation. Additionally, hardware utilization improvements could increase from 30-40% to 70-80%, further enhancing performance.
Controllability is a key focus in video model development, with the H3 Max Director being the only model capable of generating up to 60 minutes of continuous action-controlled videos. This model has gained significant popularity, with usage volume more than double that of its competitors, and it can retain details from the last two minutes of video.
Enhancements are underway for the H3 Max Director to enable seamless transitions and real-time content control, which could open new creative avenues. However, there is a noted trade-off regarding a slight quality loss with the H3 Max Turbo compared to the original model. The media market for video models is characterized by bursts of innovation, reflecting a dynamic and rapidly evolving landscape.
Collaborations with AI IP holders are underway to create live versions of shows using the H3 Max Director, incorporating voice prompting for interactive user experiences. The Fall Live experience demonstrated the model's capabilities and gained viral attention on Twitter, highlighting its potential for real-world applications.
Progress in AI models is also being made in addressing legal and data residency issues, which are becoming increasingly manageable. A new system allows individuals to apply their own intellectual property into AI models, enhancing customization and control over content.
The launch of Seedance US hosting meets a significant demand from Hollywood studios for localized AI solutions. In the near future, Hollywood studios are expected to dramatically increase their AI usage, with projections suggesting a rise of 10x to 100x.
The Genet-to-Media conference has shifted its focus to studios eager to integrate AI into their production workflows, with attendance notably increasing, reflecting a growing interest in AI technologies within the industry. The conference aims to tackle the specific challenges and needs faced by professionals in the evolving landscape of AI in media.
This summary was generated from the episode transcript and can contain mistakes.