SeedVR2 VideoUpscaler, in my opinion, is the best node for upscaling images. However, a common problem is that this node uses a very high amount of CUDA VRAM. If you want high-resolution images, you often encounter errors indicating insufficient CUDA VRAM.

Therefore, I developed this workflow. It uses a tiled technique to divide the image, upscale it into sections, and then reassemble them.

Dividing the image into rows and columns is crucial. This workflow doesn't involve many nodes and is relatively easy to understand. Therefore, I'd like to share it so everyone can try it.

** Show case executed in 256.02 seconds (4 minutes) VRAM 16GB

** 1024*1024 -> 4096*4096 | rows 4 & cols 4