Klarity is a local, free, offline AI-powered image and video restoration tool that delivers professional-grade denoising, deblurring, upscaling, and frame generation. Built for creators, editors, and anyone who needs to enhance visual content without subscription fees.
📦 Pre-built binaries available — no Python or setup needed. Grab the latest release with CPU binaries for Windows and Linux, download, extract, and run.
🤖 For AI agents and automated tools: See Bots.md
# Clone the repository
git clone https://github.com/HAKORADev/Klarity.git
cd Klarity
# Install dependencies
pip install -r requirements.txt
# Launch GUI
python src/klarity.py
# Or use CLI mode
python src/klarity.py cli# Install FFmpeg (required for video processing)
# Windows: winget install FFmpeg
# macOS: brew install ffmpeg
# Linux: sudo apt install ffmpegKlarity offers nine distinct processing modes, each designed for specific enhancement needs:
| Mode | Description | Input | Output |
|---|---|---|---|
| Denoise | Remove noise from images/videos | Image/Video | Image/Video |
| Deblur | Remove blur from images/videos | Image/Video | Image/Video |
| Upscale | 2x or 4x upscaling | Image/Video | Image/Video |
| Clean | Denoise + Deblur pipeline | Image/Video | Image/Video |
| Full | Denoise + Deblur + Upscale | Image/Video | Image/Video |
| Frame-Gen | AI frame interpolation | Video | Video (higher FPS) |
| Clean-Frame-Gen | Clean + Frame generation | Video | Video |
| Full-Frame-Gen | Full pipeline + Frame generation | Video | Video |
| Mode | Models | Download | Quality | Speed |
|---|---|---|---|---|
| Heavy (default) | NAFNet-width64, RealESRGAN-x4plus, RIFE v4.25 | ~785 MB | Best | Slower |
| Lite | NAFNet-width32, RealESRGAN-general-x4v3, RIFE v4.17 | ~204 MB | Good | Faster |
| Metric | Lite Mode | Heavy Mode |
|---|---|---|
| Hardware Demand | 🟢 13x less demanding | 🔴 Baseline |
| Processing Speed | 🟢 20x faster | 🔴 Baseline |
| Quality Loss | 🟡 ~20-28% quality trade-off | 🟢 Maximum quality |
| Minimum RAM | 🟢 4GB RAM | 🔴 8-16GB RAM |
| Best For | Quick previews, low-end systems | Final output, maximum fidelity |
💡 Pro Tip: Use Lite mode for rapid iteration and previewing, then switch to Heavy mode for your final export.
Klarity leverages state-of-the-art open-source models for professional-grade restoration:
- Denoising: NAFNet-SIDD — Neural networks for noise reduction
- Deblurring: NAFNet-GoPro — Motion blur removal
- Upscaling: Real-ESRGAN — 4x super-resolution
- Frame Generation: RIFE — AI frame interpolation
- Launch:
python src/klarity.py - Drag & drop files or click Browse
- Select processing mode from dropdown
- Choose model mode (Heavy/Lite)
- Set upscale factor (2x or 4x) if applicable
- Click "Process"
- Preview results with comparison slider
- Save output
python src/klarity.py cli# Image processing
python src/klarity.py deblur input.jpg -o output.jpg
python src/klarity.py denoise image.png
python src/klarity.py upscale photo.jpg --upscale 4
python src/klarity.py clean image.jpg # denoise + deblur
python src/klarity.py full image.jpg # full pipeline
# Video processing
python src/klarity.py frame-gen video.mp4 --multi 2
python src/klarity.py clean-frame-gen video.mp4 --multi 4
python src/klarity.py full-frame-gen video.mp4 --multi 2 --upscale 2
# Lite mode (faster)
python src/klarity.py -lite full image.jpg| Component | Minimum | Recommended |
|---|---|---|
| CPU | 2 cores | 4+ cores |
| RAM | 4GB | 16GB+ |
| GPU | None (CPU works) | NVIDIA GTX 1060+ |
| VRAM | N/A | 4GB+ |
| Storage | 2GB | SSD recommended |
Note: Klarity works on CPU. GPU with CUDA significantly speeds up processing.
.jpg, .jpeg, .png, .bmp, .tiff, .tif, .webp
.mp4, .avi, .mov, .mkv, .webm, .flv, .wmv, .m4v
Left: Original (zoomed) — Right: Cleaned with Lite mode (zoomed)
Left: Original — Right: 2x Upscale (Full-Lite mode)
Left: 4x Upscale (Full-Lite) — Right: 4x Upscale (Full-Heavy)
15FPS_x4_clean.mp4.mp4
Left: 15FPS gameplay (NFS Heat) — Right: x4 Clean Frame-Gen Heavy (60FPS) at 480p
🎮 Demo: 15FPS gameplay video transformed to smooth 60FPS using Clean-Frame-Gen Heavy mode
Note: set playback speed to 0.25 to see the difference clearly
- Guide.md - Complete technical documentation
- CHANGELOG.md - Version history and changes
- Bots.md - For AI agents and automation
Each component has its own license:
- NAFNet: Apache 2.0
- Real-ESRGAN: BSD 3-Clause
- RIFE: MIT
- Klarity: MIT
- 🤗 HuggingFace Space: Go To Demo Space
- 🧪 Google Colab: open Notebook
📝 Note: The HuggingFace demo uses Lite models and supports image processing only. For Heavy models and video support, use the desktop version or the Google Colab notebook, which provides the full experience with Heavy models and video processing — just like running it on a real PC except GUI is desktop-only.