fastblend for comfyui, and other nodes that I write for video2video. rebatch image, my openpose
fastblend node:
- smoothvideo(逐帧渲染/smooth video use each frames)
- interpolateKeyFrame(插帧、只选一部分帧渲染/smooth video only use a portion of the frames)
parameter meaning:
- accuracy: larger is better, 1 is enough for most situation
- window_size: how smooth the video.
- minimum_patch_size: odd number(important), larger is better
- num_iter: Number of iterations, larger is better
- Guide weight: a parameter that determines how much motion feature applied to the style video.
- Batch size: a larger batch size makes the program faster but requires more VRAM.
- time complexity(accuracy=1): log(window_size) * minimum_patch_size^2 * num_iter * video length * video resolution
other nodes for making video:
- rebatchimage: especially for fastblend,Simultaneously render images (720p, 24G video memory, batch_size can be adjusted to 40), speed up about 40%.
- myopenpose: 因为dw_openpose在人物转身的时候不能识别人物后背,但是在人物姿势方面比openpose准确,而openpose可以很好的识别后背,这个就是结合了它两的优点
examples are in example directory.And other nodes don't have much use,so I'm not going to introduce.
Install:download the .zip and unzip to ComfyUI\custom_nodes.
Fastblend main code is from https://github.com/Artiprocher/sd-webui-fastblend I just made it work in ComfyUI.
March 28, 2024: New node:
- alert when finished: just input the full path(......\custom_nodes\ComfyUI-fastblend\drop.wav) of a sound, it will play after this node gets images.
- merge image list: the "Image List to Image Batch" node in my example is too slow, just replace with this faster one.