It seems like every few weeks a new image generation model is announced, but it's not often that one of AI's major image production players drops one.

The company has announced its newest base model for image generation, “Frames,” which according to Runway represents “a major step forward in style control and visual fidelity.”

The focus is on integrating stylistic choices and ensuring more consistency across the model, something many models have struggled with. This will allow for more consistency across video generations that use the same “world.”

In a new blog post, the Runway team stated that they will “gradually” roll out access to Frames within Gen-3 Alpha and the Runway API. We spoke with the company and were told that this will take place over the next few weeks.

In the X thread above, you can see some impressive examples of Frames' use, from 1980s SFX-inspired makeup to 1970s album art. The “world” that Jamie Ampherson, Runway's creative director, shared on X even included everyone wearing duck heads.

Indeed, it is difficult to point to anything that tells us unequivocally that the generated models are not real people.

I was particularly impressed with the Japanese zine look. It adds a sense of style over precision that often baffles other models, and while other models are almost too stylized, Runway's Frames looks great.