dain-ncnn-vulkan VS rife-ncnn-vulkan

Lien vers les instructions [https://github.com/nihui/dain-ncnn-vulkan/blob/master/readme.md de Readme.md)

I also found this and this, which FlowFrames seems to be based on. These would probably help with the frame extrapolation and increasing, but won't help me with the upscaling.

not SD based, but maybe this helps? https://github.com/nihui/dain-ncnn-vulkan https://github.com/nihui/rife-ncnn-vulkan

I can recommend DAIN for the 60 fps: https://github.com/nihui/dain-ncnn-vulkan

link to the instructions https://github.com/nihui/dain-ncnn-vulkan/blob/master/README.md

For AMD. But it takes forever to even do a few seconds on AMD for me (5700 XT; and Nvidia GPUs are impossible to get..). And NVIDIA. Generally you split the video in single frames; interpolate them with the DAIN AI Models (Idk which model is best for animation) and the images that the model created are then re-rendered to a video with ffmpeg.

flowframes is just an interface for interpolation models, you can still use rige-ncnn directly if you are familiar with the stuff. https://github.com/nihui/rife-ncnn-vulkan

Messing around with https://github.com/nihui/rife-ncnn-vulkan

Compressing the videos has become quite a project that involves the seed_travel script, a little imagemagick, upscaling with realSR, an absolute ton of interpolation with RIFE, and the swiss army knife of video tools, ffmpeg.

not SD based, but maybe this helps? https://github.com/nihui/dain-ncnn-vulkan https://github.com/nihui/rife-ncnn-vulkan