Runway's Gen-3 Alpha is a big leap for video AI

Runway just announced Gen-3 Alpha, their latest and greatest video generation model. And boy, does it look slick! Say hello to a new era of high-quality, ultra-controllable AI videos.

What's going on here?

Runway's been hard at work training Gen-3 Alpha on a brand-new setup designed for large-scale multimodal training. The result? A major level-up in fidelity, consistency, and motion compared to their previous Gen-2 model.

What does this mean?

Runway's Gen-3 Alpha is trained on both videos and images for maximum versatility. Gen-3 Alpha will power all of its existing video and image tools with boosted performance. More features to provide fine-grained control over elements like structure, style, and movement are in the pipeline.

Runway claims Gen-3 Alpha has seamless transitions & keyframing, excels at generating expressive, realistic humans, and understands a wide range of artistic styles and cinematic lingo (hello, storytelling potential!).

Here are a few prompts they’re showcasing the results for:

• Subtle reflections of a woman on the window of a train moving at hyper-speed in a Japanese city.

• An astronaut running through an alley in Rio de Janeiro.

• An empty warehouse dynamically transformed by flora that explodes from the ground.

• Close-up shot of a living flame wisp darting through a bustling fantasy market at night.

• An FPV shot zooming through a tunnel into a vibrant underwater space.

Plus, Runway is collaborating with top media players to create custom versions tailored to their specific needs. You can reach out to Runway if you want one.

Why should I care?

Looks like the models understand our world—just like OpenAI’s Sora and the quality is great too—something Luma Labs’ Dream Machine has lacked a bit.

But right now, it’s only to get excited. Gen-3 Alpha is not out yet. Runway ships, but any wait is minus points. For now, we can just record Runway’s entry in the logbook for next-gen video AI.

When it does, Runway will release the model with its beefed-up safety measures, including a new moderation system and provenance standards. That I’m okay with. Video AI is getting to the point where people can be fooled into thinking it’s real.