Runway unveiled its Gen-3 “better and faster” artificial intelligence video model two weeks ago promising access “in the coming days." It is now available to a select group of Creative Partners Program members who’ve been sharing some incredible insights into what is possible.
Gen-3 is a “Sora-like” model, offering more natural motion, improved realism and initial video clips of 10-11 seconds, rather than the 2-3 seconds of the previous generation.
The new model was rebuilt from the ground up using a new generation infrastructure purpose-built for large-scale multimodal training.
This new model was trained on image and video at the same time for improved realism, however, during testing it only works with text-to-video.
Videos we’ve seen shared include a spaceman in a music video, a hyper-realistic woman aging as the camera moves around her, and title screens with animated words.
How well does Gen-3 work?
I’m not a Partners Program member so can’t give a first-hand account of using Gen-3, but from what those with access have said it needs image-to-video but otherwise is impressive.
The thing I’m most excited about is the fact Gen-3 even works well across lip-sync. This allows for full-body animation alongside accurate lip movement.
1. A spaceman in a music video
Runway Gen-3 Alpha is INCREDIBLE! 🤖As a Creative Partner, I have had early access for just over 24h and here are some fun visuals I have made 🧵⬇️#gen3 #Runway 🔊🆙 pic.twitter.com/ZR9GhSUn4TJune 30, 2024
Creative Partner NoBanksNearby made an impressive dance sequence-style music video featuring a spaceman and an impressive amount of bass.
In the same thread, they also shared a series of other spaceman-inspired videos all generated using Runway Gen-3. Writing on X: “I am letting my creativity run wild and having a lot of fun and we aren't even into the full release yet with [image-to-video]”.
2. Title screen animation
Wow...Runway Gen-3 excels at creating stunning title screen animations.9 impressive examples:1. pic.twitter.com/PAi80T7WNAJune 30, 2024
Being able to create legible and consistent text is something AI image tools are only just getting to grips with but it looks like Runway has it down in Gen-3.
ProperPrompter shared a series of impressive clips showing title animations including one showing the word Runway appearing, formed from surrounding leaves.
3. VFX and cinematic moments
5. "Astronaut walking in a scary spaceship, science fiction, cinematic film, H. R. Giger" pic.twitter.com/FFlt4SR0ejJune 29, 2024
Javi Lopez, the founder of Magnific, the AI upscaler & enhancer, created some impressive graphical shots including a "cinematic shot of water inside fire in a fast spiral" and a "cinematic shot of a gorgeous woman made of water".
My favorite of his has to be the astronaut walking through a scary spaceship as it gives off both VFX and AAA game vibes at the same time.
4. Zombies at the table
Wow! pic.twitter.com/jNf5Nj0oB3June 29, 2024
User Maxescu asked for prompts from the public to share and one of them was of zombies in tuxedos sitting at a table on a 1940s ocean liner drinking team.
The exact prompt was: "B&W footage of zombies in tuxedos sitting at a table on a 1940s ocean liner drinking tea," shared by Jason Wissinger along with his own Luma Dream Machine clip.
5. A monster from the deep
"a cinematic shot of a hideous monster rising from the river Thames in London" Gen 3 pic.twitter.com/X31GQLOSL7June 28, 2024
Uncanny Harry created a stunning video of a "hideous monster rising from the river Thames in London". It indicates how AI will allow even small indie films to include blockbuster-style VFX.
