ZeroHackz/OpenFlowFrames
A lean, fully free fork of Flowframes by n00mkrad, rebuilt as a modern Python (CustomTkinter) GUI around the latest RIFE models — no Patreon tiers and no paid builds. It interpolates a video file or a folder of PNG/JPG/WebP frames (with a custom input framerate) to MP4 with the original audio preserved, or exports the interpolated frames as a PNG sequence, normalizing mixed resolutions and alpha/16-bit PNGs automatically. The same engine is exposed two more ways: a headless CLI for scripts, CI and agents, and — the reason it is listed here — a DeepSeek Harness plugin bundle that gives an agent flowframes_probe and flowframes_interpolate tools so it can inspect and interpolate clips directly, including slow-motion conversions and frame-folder workflows. The README notes it pairs with the author's GUI-SAM2Matting plugin.
Install
dsh plugin add github:ZeroHackz/OpenFlowFramesREADME plugin section: install the repository as a dsh plugin bundle and an agent can interpolate video directly through two tools — flowframes_probe (inspect a video or frame directory: resolution, fps, frame count, audio) and flowframes_interpolate (run RIFE interpolation headless to MP4 with audio, or PNG frames). Requirements on the machine running dsh: Python 3.10+ (or set OPENFLOWFRAMES_PYTHON); the bundled packages/rife-ncnn model files ship with the repo so no model downloads are needed. For standalone use the README documents launcher-gui.bat (creates a venv on first run and launches the GUI), a portable build via build-portable.bat producing a self-contained dist/ folder, and a headless CLI (python cli.py --probe ..., --factor, --model, --out-mode).
Compatibility
Windows-oriented tooling (.bat launchers, portable exe). Python 3.10+ on the dsh host. Interpolation runs through rife-ncnn-vulkan with RIFE 4.9–4.26 bundled, so AMD, Intel and NVIDIA GPUs all work without CUDA or PyTorch.
Details
- Repo: ZeroHackz/OpenFlowFrames
- Category: Other
- Stars: 4
- Version: GitHub install (README documents the plugin form only; no npm package is named)
- Last push: 2026-08-13
- First seen: 2026-07-04
Recent updates
README describes the pipeline end to end: ffprobe reads input framerate and frame count, ffmpeg extracts frames (skipped for frame-folder input), rife-ncnn-vulkan interpolates to frames × factor, and ffmpeg re-encodes H.264 at the multiplied framerate while copying the original audio — or the frames export as PNGs. Credits name Flowframes (n00mkrad), RIFE (hzwer), rife-ncnn-vulkan (nihui) and FFmpeg.
FAQ
- How do I use OpenFlowFrames from DeepSeek Harness?
- Install the repository as a dsh plugin bundle (dsh plugin add github:ZeroHackz/OpenFlowFrames). The agent then gets flowframes_probe to inspect a video or frame directory and flowframes_interpolate to run RIFE interpolation headless. Python 3.10+ is required on the machine running dsh.
- Do I need an NVIDIA GPU or CUDA?
- No. Interpolation runs on rife-ncnn-vulkan with RIFE 4.9–4.26 models bundled in the repo, so it works on AMD, Intel and NVIDIA GPUs (or CPU) without CUDA or PyTorch and without downloading models.
- Can it keep the audio?
- Yes — the MP4 output re-encodes H.264 at the multiplied framerate and copies the original audio. You can also export the interpolated frames as a PNG sequence instead.