부드러운 영상 변환! AnimdateDiff+LCM (comfyui 그대로 따라하기)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Hello. NeuralNinja here. This video will show you how to quickly transform videos using Animate Diff and LCM. We'll start with basic image transformation, then apply ControlNet and try some face detailing. Most of the nodes are basic, so you can follow along easily. For running on Colab site, please check the ControlNet AUX node. Also, please enable high RAM options if possible. (Pro users) Let's start with the basic workflow. I'll add the prepared prompt. Adjust the size and create it right away. It's generated well. Let me increase the steps a bit. Now, to speed up generation, I'll apply LCM. You can easily use LoRA settings. Now, let's add LoRA and select LCM. I'll connect the nodes. I'll also add Model Sampling Discreet. It's recommended to use them together. I'll connect the nodes.(Even if I use it without it, I can't feel much of a difference.) I'll reduce the steps to 8. Please change the sampler to LCM. With more than a quarter reduction in steps, it's faster. Applying LCM like this makes image generation much faster. Let's also try Image2Image. I'll add VAE Encoder. I'll connect the input images to the K-Sampler via VAE Encoder. I'll lower the Denoising for some variance. I'll connect the nodes. You can also convert videos into images and apply the same. I'll upload the video to Google Drive and load it. (On your local computer, just paste the video file path.) I prepared a short 1-second video. I'll change Load Image to Load Video. This is a node that loads an image using a file path. I'll copy and paste the uploaded paths. (On your local computer, just paste eg. d:/work/clip.mp4) I'll lower the Denoising. I'll change it to Video Combine node. Convert multiple images into a video file and preview. Video transformation takes time as it involves many images. I'll add a Resize node to limit the size. Now, let's transform. Each frame image has been transformed. Notice the differences between images? With 8 frames, the speed isn't matching. Let's adjust the frame rate. We'll adjust it while loading the video. Let's add some ControlNet to provide uniformity. Please add Advanced. You can apply it along with intensity. Start with Open Fold ControlNet settings. I'll use DW Pose for Preprocessor. (For the DW Pose, please refer to the link in View More.) Connect the images processed in Preprocessor to ControlNet. I'll add a preview to check the processed images. I'll connect the prompt. I'll add another ControlNet. This time, I'll apply Depth ControlNet. I'll add Preprocessor. Adjusting the End percent. Finally, I'll apply Line Art ControlNet. I'll add preview images. Now, let's generate. Applying ControlNet makes it slower. But it provides uniformity in shape. Let's check the preview images. Open Pose adjusts the movements. Depth Map adjusts the depth and area. Line Art adjusts the outlines. Now, let's use Animate Diff to connect more smoothly. I'll also add Context options. I'll adjust the positions. I'll connect the nodes. I'll increase the Denoising. Now, let's generate. Using Animate Diff seems to require Advanced ControlNet nodes. I'll change it. It's definitely smoother, right? I'll adjust the ControlNet intensity. The face looks too small, so it's not well represented. Let's try correcting it with Face Detailer. like ADetailer, DDetailer transform only the face I'll apply LCM. I'll connect the nodes. I'll connect the Detector (face, person, hand) I'll apply LCM. I'll set Denoising to weak. Even with a weak setting for the face, it seems to work fine. The face is clearly sharper now. Now, let's try more longger video. This is a 5-second video. That's all for using Animate Diff and LCM to transform videos. Using Animate Diff seems to greatly improve the consistency of videos compared to before. I hope this video was helpful. We'll see you next time with another great video. Thank you.
Info
Channel: 뉴럴닌자 - AI공부
Views: 8,212
Rating: undefined out of 5
Keywords:
Id: MCG_PJyazhs
Channel Id: undefined
Length: 17min 16sec (1036 seconds)
Published: Sat Jan 27 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.