Runway Unveils Gen-3 Alpha Turbo: A Faster, More Affordable AI Video Model

Runway, the New York City-based startup, has spiced up the AI landscape with the introduction of its Gen-3 Alpha Turbo model. Just days after unveiling its innovative image-to-video capability for the Gen-3 Alpha model, the company announced via X (formerly Twitter) that Turbo would be "rolling out...with significantly lower pricing over the coming days."

Runway highlighted that the Turbo model operates at "7x faster than the original Gen-3 Alpha." Co-founder and CEO Cristóbal Valenzuela noted that users could generate 10-second videos in just 11 seconds—almost in real-time. We previously tested the earlier Gen-3 Alpha, which already generated videos fairly quickly, often within a minute. However, Runway is committed to enhancing speed and quality in the competitive field of generative AI video models, where competitors like Pika Labs, Luma AI, Kling, and OpenAI’s Sora are emerging.

Valenzuela also mentioned that Runway is working on updating its mobile app to support image-to-video with Gen-3 Alpha.

Why Offer More for Less?

The decision to launch a faster model at a lower price raises questions. It's possible that Gen-3 Alpha Turbo is a lightweight version, making it less computationally intensive for Runway’s servers and therefore cheaper to run. Moreover, faster generation could encourage increased usage, ultimately driving higher revenue through subscription plans or a la carte "credits" for video generation.

Currently, Runway provides various monthly subscription options, each with a fixed number of credits for still image or video generations. For reference, the previous model, Gen-3 Alpha, requires 10 credits per second of video, while the Gen-2 model costs 5 credits per second. Surprisingly, the oldest Gen-1 model is the most expensive, at 14 credits per second. It seems reasonable that Gen-3 Alpha Turbo might be priced around 5 to 7 credits per second.

Training Controversies Persist

Recent reports suggest that Runway may scrape popular YouTube channels, including copyrighted material, to train its models. While this tactic has stirred criticism, Runway has not officially responded to these allegations. The company is already facing lawsuits alongside other generative AI firms for purported copyright violations involving still images.

Despite the backlash, scraping content for AI training is often viewed as permissible, especially given that Google has employed similar strategies to build its search index. However, Ed Newton-Rex, a former Stability AI executive and founder of Fairly Trained, called for transparency regarding Runway's training data on X, reiterating concerns about unauthorized data usage.

As ongoing lawsuits unfold, it remains to be seen whether they will compel generative AI companies like Runway to reveal their training methodologies, particularly regarding compliance with copyright laws and potential repercussions.