Runway claims its latest text-to-video model produces even more accurate visuals than its predecessor. Runway said in a blog post on Monday that its 4.5-generation models can produce “cinematic and highly realistic output,” which can make it even more difficult to tell what’s real and what’s AI.
“Gen-4.5 achieves unprecedented physical and visual precision,” Runway’s announcement said. It adds that the new AI model is better able to follow prompts and can generate detailed scenes without compromising video quality. Runway says AI-generated objects “move with realistic weight, momentum, and force,” while liquids “flow with appropriate dynamics.”
Runway says the Gen-4.5 model is being rolled out to all users in stages and will offer the same speed and efficiency as the previous generation. However, the model still has some limitations due to potential problems with object persistence and causal inference. This means that the effect may outweigh the cause, such as a door opening before someone uses the handle.
Alongside Runway, OpenAI is ramping up its efforts to make AI-generated videos look more authentic. OpenAI highlighted physics upgrades in its Sora 2 text-to-video model released in September, with Sora head Bill Peebles saying, “You can accurately perform a backflip on a paddle board in a body of water, and all fluid dynamics and buoyancy are accurately modeled.”
Runway says its Gen-4.5 model is also better at handling different visual styles and can produce more consistent photorealistic, stylized, and cinematic visuals. The company claims that the photorealistic visuals created with Gen-4.5 are “indistinguishable from real-world footage with lifelike detail and accuracy.”
