Flowframes
for Windows
by n00mkrad
Flowframes is a Windows app that uses the RIFE neural network to generate the frames between existing ones, so 24 or 30 FPS footage plays back at 60, 120 or a custom rate. Processing runs on your own Nvidia, AMD or Intel GPU through CUDA or Vulkan. The 2.5 MB installer fetches the AI models on first run.
About Flowframes
Flowframes is a desktop front end for AI frame interpolation, the technique that invents realistic in-between frames instead of duplicating existing ones. Drop in a video, pick a multiplier such as 2x or 4x, and the app renders a new file at the higher frame rate. The output is the kind of motion smoothing a high-end TV does, but baked into the file.
The default model is RIFE, available in a CUDA build for Nvidia cards and an NCNN/Vulkan build that also runs on AMD and Intel GPUs. Other models such as DAIN and FLAVR can be added. Because everything runs locally, there is no upload, no queue and no per-video charge; render time depends on your GPU and the clip's resolution.
Animation gets special handling. 2D cartoons and anime are often drawn at 12 or 8 frames per second and then padded with duplicates, and interpolating those duplicates produces stutter. Flowframes can strip duplicate frames before interpolation and put them back afterward. Scene-change detection stops the model from morphing across a cut.
Original audio, subtitles and metadata are carried over to the output, and image sequences and GIFs work as input too. Version 1.36.0 needs Windows 10 or 11 plus a GPU with Vulkan support. The project is developed by n00mkrad and released under GPL-3.0; the code is public on GitHub.
