text-to-video: Runway GEN-2 update! (step-by-step)

Hello, amazing people! Today I'll give you a much awaited update on the state of the art for text-to-video: RunwayML GEN-2, updated!

Here's a breakdown of how it was just 1 month and 2 weeks ago (!):

See older post here.

Here's a breakdown of what we'll be testing today:

Cat running in the wild

An astronaut cat in space

A cat reflecting on the meaning of life

For each one of those, that were the same as last post btw, let's see what it's yielding now:

Cat running in the wild

4 previews, and I basically have to choose between better face and 2 legs, or worse face and 3 legs. Yikes. Let's go with #3, as it looks like a solid compromise.

Here's what we got as a video:

If you recall, last version of this was full nightmare material, so while this now is still weird af, at least you can tell it got the prompt right.

4/10, too wonky, but setting feels right.

An astronaut cat in space

Would probably choose #4 as it reminds me of our orange cat ("Bowie”), but #3 looks solid-er, so let's see what it yields.

I must admit, this one was the best BY FAR in all my text-to-video generations. Cute, does what was supposed to do, solid choice all-around. 8/10

A cat reflecting on the meaning of life

I like how #1 and #3 look like something straight out of Full Metal Alchemist, a fusion of a cat and a dog. Sigh. Let's choose #4.

Also a delightful surprise. Seriously, check the previous post and this now is waaay better. 7.5/10, blurry background.

Main takeaways:

• The tooling is still somewhat bad, but it's clearly better than it was just 40 days or so ago. That's crazy. Let's see again in 40 days or so.

• If you want to see some amazing use cases out of it, I suggest looking at this amazing video

So, what did you think?