Transform existing videos with new styles, lighting, textures and treatments using Gen-3 Alpha Video to Video. Learn how in today's Runway Academy.

Loving it!

Dear RunwayML, Please remove the unlimited account package so both you and your paying customers can relax. You've made it nearly impossible to create anything in Explore Mode. With the fee for an unlimited account, we can get 8000 credits from a competitor and produce higher-quality outputs. I strongly suggest you take this feedback into consideration, or your customers may leave you very quickly!

This is amazing and is going to lead to great progress in AI and world models!! I can imagine that this could be used to train and distill a smaller model for real time videogame enhancement. So that a basic game engine (N64 level graphics) is running control to action + logic and a finetuned version of the realtime model (based on the artistic vision of the developers) generates the final graphics. Imagine creating concept art for a game and using a larger video diffusion model prompted with those images to generate the desired video game style and use that data to finetune a Lora for the smaller realtime model. A second thing that this tech unlocks is the ability to generate a lot of video data to train a fully promptable world simulator (like the Doom diffusion project, but general). Research labs could have multiple RL agents playing many simple lookin games (N64 level) and the video data for each game could be multiplied thousands of times by these video to video models. Then the control + video data can be used to train world simulators. All in all, I´m really excited to see this. Hope you add image prompting for the video to video mode soon!!

Amazing features

When Gen3-Beta ?!

Sooooooooooo good 🙌🏼

Awesome, right on!

👀👀👀👀