openflowframes

This project is a fork of the original Flowframes by n00mkrad. Huge thanks to him for creating and maintaining the powerful core of this application!

This fork is a lean, fully free and open reimagining: a modern Python GUI around the latest RIFE models — no Patreon tiers, no paid builds, no legacy codebase.

Screenshot

Main interface: pick a video or a folder of frames, choose a model and factor, interpolate.

✨ Features

• Latest RIFE Models: RIFE 4.9 through 4.26 bundled in the repo — no external model server, fully self-contained.
• Any GPU: Runs on AMD, Intel, and NVIDIA via rife-ncnn-vulkan — no CUDA or PyTorch required.
• Modern GUI: A clean CustomTkinter dark-mode interface — no install needed beyond Python.
• Video or Frame-Folder Input: Interpolate a video file, or a directory of PNG/JPG/WebP frames with a custom input framerate.
• MP4 or PNG Output: Encode to H.264 MP4 (audio preserved) or export the interpolated frames as a PNG sequence.
• Robust Frame Handling: Mixed resolutions and alpha/16-bit PNGs (common in AI-generated frames) are normalized automatically.
• Portable Windows Executable: Build a self-contained portable app with one script — no Python needed on the target machine.
• No Monetization: Fully free; no Patreon/PayPal integrations.

💻 How to Use (Easy Way)

Python GUI (recommended):

1. Clone this repository:

   git clone https://github.com/ZeroHackz/OpenFlowFrames.git
   

2. Double-click launcher-gui.bat — it sets up a virtual environment on first run and launches the GUI.

Portable build:

Run build-portable.bat. The result in dist/ is fully self-contained:

dist/
  OpenFlowFramesPortable.exe
  packages/av/           (ffmpeg)
  packages/rife-ncnn/    (interpolator + bundled RIFE models)

Copy the dist folder anywhere and double-click the exe.

How It Works

1. ffprobe reads the input framerate and frame count (or you provide the FPS for a frame folder).
2. ffmpeg extracts frames (skipped for frame-folder input).
3. rife-ncnn-vulkan interpolates to frames × factor.
4. ffmpeg encodes H.264 at the multiplied framerate, copying the original audio — or the frames are exported as PNGs.

⌨️ Headless CLI

The same engine runs without the GUI — from scripts, CI, or agents:

python cli.py --probe --input C:\clips\dance.mp4
python cli.py --input C:\clips\dance.mp4 --factor 2 --model "RIFE 4.9" --output out.mp4
python cli.py --input C:\clips\frames --fps 24 --factor 4 --out-mode png --output out_frames

🤖 DeepSeek Harness Plugin

This repository is also a DeepSeek Harness (dsh) plugin bundle — install it into a dsh profile and an agent can interpolate video directly through two tools:

• flowframes_probe — inspect a video or frame directory (resolution, fps, frame count, audio).
• flowframes_interpolate — run RIFE interpolation headless (MP4 with audio, or PNG frames). Works on CPU or any GPU — AMD, Intel, or NVIDIA, no CUDA needed.

dsh plugin add github:ZeroHackz/OpenFlowFrames

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.

Example prompts for the agent

• Interpolate E:\clips\dance.mp4 to 60fps with RIFE 4.9 and save it as dance-60fps.mp4
• This folder of AI-generated frames is at 24fps — make a smooth 4x version, then export it as an MP4
• Probe every video in E:\clips and list their fps and frame counts
• Convert this 24fps clip to 120fps slow-motion footage and matte out the foreground afterwards (pairs with the GUI-SAM2Matting plugin)

Credits

• Flowframes by n00mkrad — the original application this fork is based on
• RIFE by hzwer
• rife-ncnn-vulkan by nihui
• FFmpeg