Today, AI video startup Runway announced the launch of its latest video synthesis model, known as Gen-4. This innovative model addresses several critical challenges faced in AI video generation, particularly the need for consistent characters and objects across different shots. If you've previously viewed short films created with AI, you may have noticed that these videos often resemble dream-like sequences, lacking a coherent narrative. Runway claims that Gen-4 can maintain character and object consistency, provided users offer a single reference image as part of their project within Runway's user-friendly interface.
Runway has published demonstration videos showcasing how Gen-4 can effectively maintain the same character—such as a woman—appearing across various scenarios and environments, as well as identical objects, like a statue, that look consistent despite being depicted in different contexts and lighting conditions. This enhancement allows filmmakers utilizing Runway's tool to capture the same environment or subject from multiple angles within a single sequence, a feat that was nearly impossible with previous models like Gen-2 and Gen-3.
The last significant update from Runway was Gen-3, launched in June 2024, which greatly increased the maximum video length users could produce from just two seconds to ten seconds. This update also improved the consistency and coherence of generated videos compared to its predecessor, Gen-2. The introduction of Gen-4 is poised to further elevate the capabilities of AI video synthesis.
Runway first introduced its publicly available video synthesis product in February 2023. The initial Gen-1 creations were largely seen as novelties rather than tools for creative professionals. However, subsequent enhancements have made the platform increasingly useful for real-world projects. Notably, its capabilities were utilized in the acclaimed film Everything Everywhere All At Once, where AI-generated sequences featuring two rocks with googly eyes conversing on a cliff were produced. The tool has also contributed to comedic segments on The Late Show with Stephen Colbert.
While many competing startups in the AI space were founded by AI researchers or Silicon Valley entrepreneurs, Runway was established in 2018 by art students from New York University's Tisch School of the Arts. The founders—Cristóbal Valenzuela and Alejandro Matamala from Chile, along with Anastasis Germanidis from Greece—pioneered one of the first usable video-generation tools available to the public and played a crucial role in developing the Stable Diffusion model.
Despite being significantly outspent by competitors like OpenAI, Runway aims to carve out a niche similar to Adobe by focusing on creative professionals such as designers and filmmakers. The company has strategically positioned itself as a support tool within existing creative workflows rather than as a standalone solution. This approach has enabled Runway to secure a partnership with Lionsgate, allowing the company to legally train its models on Lionsgate's extensive film library while providing tailored tools for production and post-production.
However, Runway has faced scrutiny, alongside other AI companies like Midjourney, concerning a widely publicized intellectual property case initiated by artists who allege that their work was used without permission for model training. As a result, not all creatives are fully supportive of the platform.
The announcement of Gen-4 coincides with its rollout to all paid plans and Enterprise customers. However, initial access may be gradual; upon logging into my paid account, Gen-4 was listed but marked "Soon," indicating it isn't yet available for use. To manage server load, Runway may be rolling out the model to accounts incrementally. Once fully available, Gen-4 will be accessible exclusively to users on paid plans, starting at $15 per month for individual, non-enterprise accounts and going up to $95 per month, with a 20% discount for annual subscriptions. Enterprise accounts are priced at $1,500 per year.
Each plan provides users with up to 2,250 credits per month, but due to the curated nature of generating usable AI videos, this number may not yield many final products. The $95 per month individual plan includes an Explore Mode that allows for unlimited generations at a relaxed rate, providing users the opportunity to experiment and refine their outputs before investing in final projects.
The improvements claimed with Gen-4 aim to address the common complaints of creatives regarding the limitations of existing video synthesis tools, particularly their inconsistency and lack of scene comprehension. Competing offerings, like OpenAI's Sora, have attempted to tackle these issues but with limited success. As the rollout progresses, the creative community eagerly anticipates how Gen-4 will enhance the landscape of AI video generation, enabling more coherent storytelling and artistic expression in the digital realm.
