A fully portable, zero-dependency multimedia player integrating MPV, an embedded VapourSynth environment, and NVIDIA TensorRT hardware acceleration to deliver real-time AI video frame interpolation using cutting-edge RIFE neural network models.
Important
This project is currently tailored exclusively for the following environment:
- Operating System: Windows 10 / 11 (64-bit).
- Graphics Card: NVIDIA RTX GPUs only (RTX 2000, 3000, 4000, 5000 series or newer — xx70 tier and above such as RTX 3070 / 4070 / 5070 recommended for an optimal, stutter-free experience). Legacy GTX series cards are unsupported due to the lack of dedicated Tensor Cores.
- Supported Resolution: 1080p (Full HD) and below only. Ultra-high resolutions like 2K (1440p) and 4K (2160p) are automatically bypassed (played natively without AI interpolation) to prevent VRAM saturation and severe rendering bottlenecks.
- Display Driver: NVIDIA Driver version 580.00 or higher for CUDA 13 and TensorRT 10 runtime support (automatically verified by
setup.bat).
- Flexible AI Frame Multiplication: Multiplies source video frame rates dynamically (2x, 3x, 4x, 5x, 6x) in FP16 precision using TensorRT acceleration, achieving cinematic fluidity based directly on the original video framerate.
- 100% Isolated & Portable: Bundled Python and VapourSynth runtime inside the project directory — zero system pollution and no manual environment configuration required.
- Modern Interface (ModernZ UI): Sleek, glassmorphic dark-purple bottom control bar with full mouse and touch-screen navigation.
- Interactive In-Screen Settings Sidebar: Adjust RIFE models, multipliers, hardware streams, video shaders, fonts, and interface language live on the fly without interrupting video playback.
- Full Bilingual Support (English / العربية): The entire interface, control bar, and settings sidebar natively support both English and Arabic with one-click live language switching.
- Real-Time FPS HUD: Native in-frame render counter tracking actual display refresh rate and frame drops without relying on third-party utilities like RTSS.
- Streaming & Media Center Bridges: Seamless one-click external player integration for Stremio and a zero-conflict runtime bridge for Harbor.
- Setup & Initialization:
- Download or clone this repository.
- Run
setup.batonce. The script will automatically verify your GPU driver compatibility, download required MPV and TensorRT runtimes, and compile the default Model 4 engine.
⏳ Important Note: During the final setup step, an automated compiler window will open to assemble the TensorRT engine for your specific GPU. This process typically takes 1 to 3 minutes and runs only once. Please allow it to complete and close automatically.
- Launch:
- Open
mpv/mpv.exedirectly or drag and drop any media file onto it to start enjoying ultra-smooth playback.
- Open
F1Key: Instant toggle to enable or disable RIFE frame interpolation on the fly, accompanied by an on-screen display (OSD) status indicator.
Open the floating settings panel by clicking the Settings icon (⚙️) in the ModernZ bottom bar. It allows live tuning of all playback parameters without pausing playback:
- RIFE Engine & Hardware Streams: Select active RIFE model, frame rate multiplier (2x to 6x), GPU execution streams, and 4K/UHD bypass behavior.
💡 Important Multiplier Tip: It is strongly recommended not to exceed 3x. Setting higher multipliers (such as 4x or 6x) introduces immense computational load that even high-end graphics cards cannot process smoothly in real-time, which will cause severe stuttering and frame drops instead of smooth motion.
- Subtitles & Typography:
- Pre-loaded with high-grade multilingual and Arabic fonts in
mpv/portable_config/fonts/(XB Zar, Cairo, Roboto, OpenSans, etc.). - Custom Font Auto-Detection: Drop any
.ttfor.otffont file into thefonts/directory, and it will immediately appear in the sidebar options for live font, size, color, outline, and position adjustments.
- Pre-loaded with high-grade multilingual and Arabic fonts in
- Video Shaders & Enhancements: Toggle Anime4K, FSRCNNX, CAS, and Adaptive sharpeners individually or clear all active shaders with a single click.
- Real-Time FPS HUD: Toggle the on-screen frame rate counter (disabled by default) and choose its display anchor (top-right, top-left, top-center).
- Language & UI Scaling: Instant switching between English and Arabic layouts, plus UI element and font scaling adjustments.
The project includes four distinct RIFE models optimized for different GPU performance tiers:
| Model # | Model Name | Description & Best Use Case | Recommended GPU Tier |
|---|---|---|---|
| 1 | RIFE v4.25 | Highest optical precision and edge clarity for high-bitrate 1080p movies | RTX 3090 / 4080 / 5080 and above |
| 2 | RIFE v4.25 Lite | Exceptional detail with optimized memory footprint and balanced throughput | RTX 3080 / 4070 / 5070 |
| 3 | RIFE v4.22 Lite | Maximum consistency and artifact suppression for anime and television series | RTX 2080 / 3070 / 4070 |
| 4 | RIFE v4.17 Lite | Ultra-lightweight and fastest processing (Pre-compiled default) | All NVIDIA RTX GPUs |
A curated collection of industry-standard GLSL and Hook shaders for live image upscaling and clarity enhancement, accessible under Sidebar (Settings -> Shaders):
- Anime4K Restore CNN (M / VL): Neural networks engineered to restore crisp linework and eliminate compression artifacts.
- Anime4K Upscale CNN x2: High-fidelity 2× neural super-resolution for animated content.
- Anime4K Darken / Thin Lines: Deepens ink outlines or thins heavy strokes for a crisp, refined modern aesthetic.
- FSRCNNX x2 (16-0-4-1): Deep 16-layer convolutional neural network for detailed video upscaling.
- FSRCNNX x2 (8-0-4-1 LineArt): Lightweight upscaler optimized specifically for line-art clarity.
- RAVU Zoom AR R3: Rapid accurate video upscaling equipped with anti-ringing compute shaders.
- AMD FidelityFX CAS: Industry-standard Contrast-Adaptive Sharpening delivering crisp textures with zero halos.
- AMD FSR RCAS: Targeted luminance sharpening algorithm derived from AMD FSR 1.0.
- Adaptive Sharpen: Smart edge-masking sharpener that enhances contours while preserving flat gradient areas.
💡 Custom Shaders Auto-Discovery: Simply drop any third-party
.glslor.hookshader intompv/portable_config/shaders/, and MPV Flow will automatically register it inside the sidebar menu for one-click toggling.
Base AI models are stored in universal ONNX format. To pre-compile engines and eliminate runtime compilation delays, standalone batch tools are provided in the tools/ directory:
build_model_1.batthroughbuild_model_4.bat: Compiles 1080p and 720p TensorRT engines for a specific model.build_all_engines.bat: Compiles all four models sequentially for instant in-player model switching.
- Navigate to
stremio/and executelink_with_stremio.bat. - MPV Flow will register as Stremio's direct external player, launching automatically in fullscreen with full RIFE AI interpolation.
- To revert: run
unlink_from_stremio.bat.
- Navigate to
harbor/and executelink_with_harbor.bat. - Links the portable VapourSynth runtime into Harbor and copies the required configuration snippet to your clipboard to paste into Harbor's Advanced settings.
- To revert: run
unlink_from_harbor.bat.
mpv-flow/
├── harbor/ # Harbor integration tools
│ ├── harbor_bridge.py # Runtime linker & bridge logic
│ ├── link_with_harbor.bat # One-click Harbor link script
│ └── unlink_from_harbor.bat # Unlink script
├── mpv/ # MPV core and configuration
│ └── portable_config/ # Isolated portable configuration
│ ├── fonts/ # Bundled interface and subtitle fonts
│ ├── script-opts/ # Interface script configurations (ModernZ, etc.)
│ ├── scripts/ # Lua scripts (Settings sidebar, FPS HUD, etc.)
│ ├── shaders/ # Curated video shaders (Anime4K, FSR, CAS)
│ ├── vs/ # Frame interpolation script (interpolation.vpy)
│ ├── input.conf # Keybinding mappings
│ ├── mpv.conf # Core MPV rendering parameters
│ └── player_settings.conf # Unified live configuration file
├── stremio/ # Stremio external player integration
│ ├── link_with_stremio.bat # One-click Stremio link script
│ ├── stremio_bridge.py # Server.js bridge engine
│ └── unlink_from_stremio.bat # Unlink script
├── tools/ # TensorRT compilation utilities
│ ├── build_model_1.bat # RIFE v4.25 engine compiler
│ ├── build_model_2.bat # RIFE v4.25 Lite engine compiler
│ ├── build_model_3.bat # RIFE v4.22 Lite engine compiler
│ ├── build_model_4.bat # RIFE v4.17 Lite engine compiler
│ ├── build_all_engines.bat # Batch compiler for all four models
│ └── build_all_engines.py # Python trtexec / vsmlrt build engine
├── vs/ # Isolated portable Python & VapourSynth runtime
│ ├── Lib/site-packages/ # Python packages (VapourSynth, vsmlrt)
│ ├── vs-coreplugins/ # Core VapourSynth plugins
│ ├── vs-plugins/ # Video filters and RIFE base ONNX models
│ ├── python.exe # Embedded portable Python 3.12 interpreter
│ └── 7z.exe # Portable archive utility
├── .gitignore # Git exclusion rules for binary and hardware caches
├── README.md # Official English project documentation
├── README_AR.md # Official Arabic project documentation
├── setup.bat # One-click automated setup and deployment script
└── setup.py # Automated dependency installer & hardware tuner
This project is built upon and inspired by exceptional open-source contributions:
- We0M/realtime-RIFE-portable: The foundational inspiration and architectural reference for the portable standalone structure and initial RIFE + MPV integration.
- mpv player: The gold standard of highly extensible, high-precision open-source media players.
- zhongfly/mpv-winbuild: For providing modern Windows binary builds with VapourSynth integration and active Git updates.
- AmusementClub/vs-mlrt: The brilliant machine learning inference plugin bridging NVIDIA TensorRT to VapourSynth.
- hzwer/Practical-RIFE: The research and development team behind the state-of-the-art RIFE (Real-Time Intermediate Flow Estimation) optical flow algorithms.
- Fredrik Mellbin / VapourSynth: The modern, high-performance video processing framework.
- ModernZ UI (Samillion): For the sleek, touch-friendly purple On-Screen Controller.
- Bloc97 / Anime4K: For the high-speed real-time anime upscaling and line enhancement shaders.
- bjin / mpv-prescalers: For FSRCNNX and RAVU neural convolutional scalers.