FFmpeg Adds GPU Frame Interpolation: 24fps Video Becomes 48fps on RTX 30 Series and Above

FFmpeg's main development branch merged a new filter called FRUC on August 29. It calls dedicated silicon on NVIDIA GPUs to synthesize intermediate frames into existing video — a 30fps clip becomes 60fps, 24fps film material becomes 48fps, with GPU hardware acceleration and virtually zero CPU load.

💡 What You Will Learn

FFmpeg's main development branch merged a new filter called FRUC on August 29. It calls dedicated silicon on NVIDIA GPUs to synthesize intermediate frames into existing video — a 30fps clip becomes 60

What FRUC actually does: take two adjacent frames, have the GPU estimate how every object in the scene moves between them, then composite a brand new frame in between. The whole pipeline uses NVIDIA's optical flow accelerator, a piece of hardware that has existed since the RTX 20 series. Important distinction: this FRUC has nothing to do with DLSS Frame Generation. DLSS FG needs engine cooperation and does real-time rendering-time interpolation; FRUC targets already-encoded video files and works offline — any video can be interpolated. Which GPUs work: the path goes through Vulkan, using Vulkan's optical flow extensions and the NVIDIA Optical Flow SDK 5.0; the minimum requirement is Ampere — which on consumer cards means RTX 30, 40, and 50 series. RTX 40 series owners get a hidden bonus: NVIDIA's official numbers put Ada's optical flow engine throughput at roughly 2.5x Ampere's, with about 15% better interpolation quality — same feature, noticeably faster and cleaner on RTX 40+. Why NVIDIA blocked it for three years (and what changed this time): the project actually tried this in 2023 and pulled off real-time interpolation, but FFmpeg rejected the patch. The reason wasn't technical, it was licensing: that version depended on an NVIDIA software library that couldn't be freely distributed, and FFmpeg as an open-source project couldn't ship it. The developers themselves said 'impractical due to licensing'. What changed this time is the implementation: the new path only uses Vulkan to pull motion vectors the GPU has already computed, and intermediate-frame compositing is done by a shader FFmpeg wrote itself — the restricted NVIDIA library is never touched. The licensing obstacle is gone, and the code made it into mainline. The same feature, three years later, finally got in by taking a different road. Cold water: this currently lives only on FFmpeg's main development branch and is not in any stable release. Users on the stable build won't see FRUC today — to try it you have to compile a development build yourself, or wait for the next major release. The same weekend saw a larger batch of NVIDIA-related changes — RTX video super resolution, HDR, dynamic vividness, Smooth Motion — appear briefly in FFmpeg and then be reverted; only FRUC survived. When it ships to general users depends on the version cadence. For video creators this is a feature worth watching. In the past, video frame interpolation either meant slow CPU-based software or installing third-party tools. Now the canonical open-source transcoder ships GPU-accelerated interpolation built in — RTX 30 series and above effectively gain a free capability.