They Promised Consistent Animation! This is Animate Anyone

Hello friends a user in my Discord kingb just posted a pretty cool thing some researchers have made a way to animate anyone and it looks pretty fantastic let's quickly check it out oh and I also met a woman who only has one leg she calls herself Eileen AI so this is animate anyone consistent and controllable image video synthesis for character animation and while looking at this it just looks like any old video from you know whatever is stock photo site but it's actually an image input which is animated through a skeleton so all these start off just with an image and that is mind-blowingly amazing we're going to look at some of the details in a bit uh first off we're just going to check out some of the videos we're going to let's look at it here on YouTube so here's going to be some examples oh it's so 720 let's use cinem mode here so here we see the the thing we saw previously and it says here R method can transform character images into animated videos controlled by desired post sequences so here they have a reference image up the top left one here and have a post sequence and that turns into video clip through all these cute little steps here that 99% of us don't really care about how it's done a few of you Kudos you guys love that stuff but most of us we just want to see the input and the output and what what's kind of cool about this is the second part here generate video of any length and three consistent and smooth I mean that's pretty amazing compared to the stuff that we have now and here they had some comparisons with the other methods and again here's the reference image and here's the driving post uh there's is the one there and this is comparison with dream pose and BDM you can clearly see if you go back here how how much more detailed and and higher rest the right one is here and just having that from one reference image it even keeps can you see look at the pattern look at the pattern the pattern stays and here's a comparison for human Dance Generation and they have a Tik Tok data set so the Disco example here is very flickery and what we're used to and the other ones on the right side here super smooth super consistent I mean just it looks like any old video it looks very real now this is obviously not super high resolution this is just a 720 so it's a little pixely we can't get around that but what we can see looks pretty amazing we have let's look at that guy walking towards the camera this looks great I'm just surprised like all the little details on the clothes everything's I haven't seen anything like this I mean unless you know this was like a generated I don't think like from a quick glance and maybe you know seeing it in your feed and maybe it's low as whatever it's going to be passed off as genuine so we doesn't just do realistic art it realism it does anime and animated stuff as well are you as blown away as I am I mean we can see stuff like this with stable Fusion like there's some there's animate diff there's warp BB synth deorum you can put a lot of those tools together but just using one image here and this triing pose this little skeleton rig here and getting this output and with this consistency there's no flicker not anything in all these Styles this is truly going to change the game now you might be asking okay Seb you sold me where can I get it well you can't that's the thing with a lot of these tools that we get to see that look super cool they are research papers and uh we have no idea where we get to see them or when but since we know this Tech is out there and is available we'll get to it somewhere I mean just remember back in the day when stab Fusion was released and you could barely create a you know good-looking still image look at how how far we we've come and you're probably going to get to this in a very short amount of time I mean the patterns I'm I'm still Blown Away by the patterns I had such a hard time working with patterns and on clothes and stable diffusion in general keeping consistency so they have a research paper which is it's not very long but it's you know it's semi longish and if you don't want to spend your time reading that I asked chat GPT to make you a summary so I'm just quickly going to read you that summary now anime 101 is a new method for turning character images into lifel like videos it uses Advanced Techniques to keep details consistent and movements Smooth by combining different models like reference net for details a post guider for control and a temporal layer for smooth transitions it outperforms other methods in creating realistic character animations the approach succeeds in fashion and dance video benchmarks though it faces challenges in handling certain movements and efficiency it marks a significant step in making character animations more believable and adaptable for various uses thank you GPT find the links in the description below as always have a good one see you

Channel: Sebastian Kamph

Views: 26,154

Keywords:

Id: LK12oC2N87g

Length: 5min 14sec (314 seconds)

Published: Thu Nov 30 2023