Transcript of Runway Act-Two vs Wan 2.1 Vace Canny and Pose Control

Hey, just a quick video here on Runway's new act two and how we could get close to it using the one 2.1 model vase with the cany and pose control. First of all, um let's have a look at act two. Act two. Oh yeah. Oh. Oh, really? Wait. And then these here are my favorite beans. Best brand. What's going on here? And then I get the call from Doc. He was upset. Somebody saw the ship. They put it on the internet. Okay. Want more? W. What are you doing? This is funfall. Whoa. Hey, are you actually going to stop me here? I've missed you ever so I heard in the bathroom was like screaming, "Help me." I'm like, "Oh my god, you don't want it to miss." Oh, sure, sure. But both hoist lights on the runway help center page. Um just under that video, it's got all of the settings there. Um so we'll just bypass all of that. Go down to here and have a look at this cuz this is your input. So the thing is it can do an image input and it can do a video input. So that's pretty impressive. Um so you can see the zoom happening in this video here and it takes the driving video and it transfers that to a still image and to a video. So that's something that if I was doing that in one or conf UI, I'm going to have to probably use like recam master or something to continue the zoom through. Um, so I'm not too sure how I'd achieve that. Um, but yeah, the other settings here. So character image, gesture guide, and you can have it on or off. So, it's a very impressive um thing. It doesn't say too much about the lip syncing though. So, it's got facial expression um setting there, but it doesn't mention anything about lip syncing. So, I'm guessing it sort of takes some of your mouth movements. Um but I don't know how well that does. So, anyway, let's look at in compi. And first of all, I'm going to go into flux context because I need my still image. So, um, I've got a a video that I've done and that's my first, uh, still from the video. And what I want to say is change the background to a snowy mountain and put snow on the man's hair and jacket, but keep all the man's body and face structure the same. So, I've done that. And when I ran that, I came with this picture here. Um, pretty much looks the same. Did what I wanted. So, now that we've got that, we can get out. Now we're in video to video vase image reference workflow. Um, got my photo that I've just uploaded. Here's my driving video. So, it's just me uh for 4 seconds turning my head there. And up the top here on this side, I'm using the GGUF. I'm using a Q5. Um, I'll have the link in the comments there, which is this one here, where you can choose one that suits you. Um, I've also use this here where I can adjust my virtual RAM there to help as well. Got sage attention on just to speed it up and cause vid on there. There's another speed up. Um, the shift is at eight. And now my prompt is a midshot of a man turning his head on a snowy mountain while it's snowing. Um, just keep it really simple there. Um, make sure that you got the right resolution there. 832x 480. And there's just the samplers at 6 one and uh uh C FG and D noiseis at one. Just keeping everything low because of that cause. And that's my output there. It came out pretty good. So yeah, no need for motion tracking anymore. We can pretty much get these things really good now. This was with the canny node over here. So this canny node sort of takes the edges of this video and it grabs that image and sort of bakes it into that video there. So in this situation of the same person getting transferred almost everything just the background and that it this is where the Kenny mode is really good. But for other things, we probably want to use the pose model. So, I'm just going to look at this. If I change my prompt over here, a midshot of a lion turning its head into a snowy mountain, and I want to use this uh line over here. Here's my line here. And then when I pressed run, my output was this this cany here. It sort of looks a bit weird when it turns. So face on it's okay, but then it sort of gets a little bit odd. So I'm in the pose workflow now. So this is exactly the same as the Kenny one except for the open pose node here. That's the only difference. It's still got the same driving video, the same image, the same prompt, but when I ran it, it came up with this really great result. Um, and yeah, so this this is what's better when you have some other sort of image that's a lot more different dimension than the original. This will work with hands and face just like the uh runways act 2. Um the lips is probably a little bit like runway 2. It might not be right on. So you at least with this you can use multi-torque to get the lips better and you could use recam master to get the the sort of camera angles that you want. But yeah, this is still okay. Um I'll have another video coming out on how to get a better quality out of this um with a thing called Puzza. Um, yeah.