A few weeks ago we reported on Runway Gen-2, an online text-to-video tool, which generates 4-second clips based on your prompts. Up until now, the free service has only been available in a closed beta that operated via a series of private channels on Runway's Discord server. However, as of today, Gen-2 is available to everyone and you can use it via the company's website instead of Discord.
To use Runway Gen-2, you'll need to create a free account and at Runway's site and then navigate to app.runwayml.com/ai-tools/gen-2 for the tool (or click the link in your control panel there). The company has had its Gen-1 tool, which modifies existing videos to make them look different (ex: turning people into Claymation), available to everyone for months, but Gen-2 takes simple text prompts such as "robot drinking a beer" and turns them into clips.
As we noted in our prior story on Gen-2, the video clips have no sound at all and they often have very limited movement. Though you can download the clips in MP4 format, they often look more like animated GIFs. For viewing purposes, I converted all the clips embedded in this article into GIFs, which has caused the colors to dither, but it seems to have no effect on their smoothness.
You can also upload a photo and then give Gen-2 a prompt that tells it to use that image as inspiration. I uploaded a headshot of myself and asked it show "this guy drinking a beer."
The person in the video looks quite a bit like me -- he's a middle-aged, bald white guy with a salt-and-pepper beard and glowing green glasses -- but he's not really drinking a beer; it's not quite clear what he's supposed to be sticking in his mouth.
After you've entered your prompt (and possibly photo), you click the Generate button and the tool tells you if you're in the queue for your request to be processed. It then gives you a percentage showing you how much of your video is completed. In my tests, it always took less than a minute between when I entered my prompt and when the video was completed. This is consistent with what I experienced when using the beta version of Gen-2, except, in that case, I'd send my prompt to a Discord bot which didn't show a progress percentage but instead would just send me back the video after a minute or so.
When the video is complete, you will see it on-screen and you can play it right there or click a button to download it as an MP4 file. It will also be stored in the assets section of your account.
Unlike with the beta version of Gen-2, it appears that there's a limit to how many videos you can generate for free. My free account showed a limit of 60 seconds of video, which at 4 seconds per clip, amounts to 15 clips.
If you use up your credits or you want extra features such as an upscaled (higher resolution) or videos that don't contain Runway's watermark, you need to pay for a Standard account which is $15 a month or $144 a year when paid at once. For that price, you get 125 seconds of video per month and can pay extra for more.
Since gaining access to the private beta a few weeks ago, I've spent a lot of time playing with Gen-2. I'm impressed with how good many of the clips are, but it's very inconsistent and people can look like they come straight from the uncanny valley, with strange expressions in their eyes or misshapen body parts.
For better or worse, the tool is pretty bad at recreating specific characters. I asked it, at various times, to create videos of the cartoon characters Peppa Pig, Paddington Bear and Blue from Blue's Clues and they look sort-of like those characters and are usually cartoony, but the likenesses aren't very accurate. One particular image of a photorealistic Peppa Pig is really creepy and seems like it was ripped from a horror film.
You can also forget about asking for something involving a corporate logo. I asked it several times to create videos that used the Tom's Hardware name or logo and got gibberish letters instead.
However, one thing Runway Gen-2 is really good at is generating images of robots drinking alcoholic beverages or doing other "bar" things. Every time I asked it for robots drinking or smoking or dancing or pouring beer or washing their hands in the men's room at a bar, I got pretty decent output. I even got good clips of robots doing standup comedy or lounge singing. Asking for robots playing billiards or foosball or darts a bar was more of a mixed bag.
Other areas of success I found include families eating meals together and time-lapse views of cities or nature scenes.
Despite the limitations of Runway Gen-2, we have seen a number of very creative people turn them into longer movies by stitching many clips together. Perhaps the most famous example of a Runway Gen-2 movie is the Pepperoni Hugspot Commercial pizza commercial, made by a creator who goes by the name Pizza Later. Pizza later worked with a friend to create a commercial for a fake flatulence drug called FlatuLess, which we've embedded below.
Note that the music and voiceovers were created with other AI tools and all the clips were stitched together by humans using Adobe After Effects. The clips look like they are from an old commercial because Pizza Later used the Red Giant VHS filter on them.
If you want to try Runway Gen-2 and see what you can create with it, all you need to do is head to runwayml.com and sign up for a free account.
