ComfyUI Pose To Video Animate Anyone Workflow! 2 of 3

welcome back to my channel in the last video we created a pose video we're going to use today to animate using animate anyone so let's get started we're going to do the load video VHS upload and we're going to use the video I created last time there you go I'm going to upscale this image join that to there I'm going to create it width is 512 that's fine animate anyone is trained on an image that's 512 wide 768 deep high and we're going to load a pose guide here and going to put the image here uh we need to load a pose guider animate anyone as you can see going to use the default weights and join that up there so we got this and we got this now when you do anime anyone sampler join the latent to the latent and we keep the seed as normal everything looks okay don't worry about this we're not going to use this yeah that looks good so we need uh so we need an image that we're going to animate load image load image so I could use this going to upscale just going to copy this because we're just going to use the same features same parameters I should say flip Vision encode here we're going to join this to this and this to this okay so we need a clip Vision we do load clip vision and join this up to there I'm going to use the P torch bin because that's what animat anyone recommends well that's in the sample image vae encode going to join this to this and load vae over here is that the animate anyone py torch diffusion py torch yep there you go there and we're going to join this to this so we got this over here separate to this and this over there which is separate to that now we need a dening unit and the reference unit load unit 3D and 2D we got 2D here load unit and a 3D here we're going to join this one this and we're going to join this one to this these are normal weights they're the ones in the animate anyone workflow we're going to use these unless you got your own uh trained weights and you want to use them you can do that you click on there you can see what they are they are a path for that one and the model path is for that one and it's um pth file same with this pth file above there you got pth file and that is the path so once you got that you got your latent you want to vae decode to there and where's our vae loader up there we're going to join this to this and Here video combine here to there and we're going to use the save image extended over here so this is going to generate a video 30 frames per a second animate diff call it a POS to video and we're going to make that into h264 10 that's fine everything looks good give that any name you like but here depending on the size of your video for example this this depending on the size of the video and depending on how much vram you have this is a good idea so I'm going to call this pose to video images and change this pH to video images there you go and now when I run this it's going to take each frame from this and animate this guy here I'm not going to do the full whack because I might not have uh enough video R we're going to skip zero do 100 so it's going to do this and Q prompt this is what it's going to do using 20 steps here so it's going to take you can find out how long roughly it's going to take after he does the first step so about 20 so 10 minutes or 100 frames so you got 100 frames there so the model is trained to animate anyone P torch I guess is trained for 512 by 768 this video is wides screen you want to change it to 512x 768 otherwise you get corrupted output and this image usually doesn't matter because you're upscaling it or down scaling it to 512 768 and it's just going to take what whatever is in in that that took uh it's still not finished that took 10 minutes and 50 seconds to do 100 frames and that's the video we got from it that's my logo and it's converted it into a video so now we've done the first 100 frames with 16 GB of vram I can do roughly 250 frames per go this this one is 324 frames so I can skip the first 100 because I've already done that and take that to zero and you'll do the rest so here it's got every frame and when it saves it in the folder 0 01 to 100 so when I run it from skipping 100 it's going to start at 101 and carry on till the end of the video let's run this this time is going to take a bit more than 10 minutes and 50 seconds here it says 9 minute and 50 seconds because it was just that a sampler process animate anyone sampler process that took 9 minutes 50 seconds the rest of the workflow took about minute that took just under 24 minutes I zoom in here you can see all the 223 frames that it created there's the rest of the video if your video is already 512 by 768 then you can bypass the upscale image same for the image file as well you can about with the settings for the animate anyone sampler to fine-tune the output for your use case in part three we will take these images and increase the detail for a better quality output like always links in the description for the workflow and GitHub repositories for the nodes used in this video If you enjoyed this please hit the like button and subscribe for more

Channel: Grafting Rayman

Views: 2,831

Keywords: comfyui, comfyui tutorial, dwpose stable diffusion, dwpose, vhs video combine, save image extended, stable diffusion, stability ai, background removal, comfyui workflow, animate anyone, pose to video, workflow, tutorial, guide

Id: fjSRNJ1BxJk

Length: 7min 3sec (423 seconds)

Published: Fri Mar 29 2024