Runway introduces Gen-4.5 Image to Video

Runway introduced Gen-4.5 Image to Video today. Here’s the intro video:

I’ll keep this blog post brief because the model suffers from the same issues I’ve explained for the last two years ever since Sora was introduced.

The intro video looks impressive, if you’re looking at it on your phone. On a laptop, computer monitor or TV not so much. The bigger the screen is the more unwatchable AI videos are. Pause the video at almost any point and the illusion of consistency and quality falls apart. Here’s a quick rundown of the sloptastic AI errors in the intro video.

Gibberish writing on store fronts and nonsense interior vehicle design.

More gibberish text. How are we supposed to enjoy a film if there are so many distractions?

The scuba diver’s face is mashed up and his hand is caught in the door instead of holding the door handle.

Lots of bad morphing and distortion.

The level of slop in this fake anime is an insult not only to anime but also to generative AI.

Because prompting alone creates a lot of slop they’ve also shown an image to video workflow starting with images generated by Google’s Nano Banana Pro model, but the lack of consistency in the character and set makes it unsuitable for professional production. You have to use photo to video, highly detailed sketch to video, or live action video to video in order to maintain some semblance and consistency from one clip to another. AI alone has too many downsides.