| FunGen | |
|---|---|
| Developer | ack00gar (GitHub) |
| License | Non-commercial |
| Platform | Windows, Linux (Python) |
| Type | AI-powered funscript generator |
| Website | GitHub |
| Based on | YOLO object detection, FFmpeg |
FunGen is an AI-powered, Python-based tool for automatically generating funscript files from video content using computer vision (specifically, the YOLO object detection framework). It aims to dramatically reduce the time required to create scripts by automating the motion detection and timing process.
Overview §
Traditional funscript creation is a time-intensive manual process: a scripter watches video frame-by-frame and places action points at precise moments. FunGen replaces this with automated analysis, detecting motion cues in video and generating corresponding funscript action sequences.
The tool includes:
- YOLO-based tracking — motion and object detection adapted for the scripting task
- FFmpeg integration — for video decoding and frame extraction
- Modular filter pipeline — post-processing filters to clean up generated scripts:
- Autotune (Savitzky-Golay smoothing)
- Amplify / clamp / invert
- Resample
- RDP simplification (remove redundant points)
- Speed limiter (device safety)
- Ultimate Autotune — applied automatically after generation
- GUI and CLI modes — a graphical interface for single videos and a command line for batch/folder processing
Performance §
FunGen achieves faster-than-realtime processing on capable hardware. Benchmarks have shown:
- ~60–110 FPS for 8K 8-bit VR video (single-process)
- ~160–190 FPS in parallel mode
A full 8K VR scene (typically 30–60 minutes) can be processed in 20–30 minutes.
Limitations §
The quality of AI-generated scripts typically falls short of expert manual scripting, particularly for nuanced timing, multi-axis data, and scenes with unusual camera angles. Generated scripts are generally used as a starting point for manual refinement in OpenFunscripter.
FunGen is licensed for personal, non-commercial use only.