7 local AI tools built into FrameShift.
No upload. No cloud. No subscription.
Every AI feature in FrameShift runs entirely on your Windows machine using ONNX Runtime. Remove backgrounds, upscale images, erase objects, split audio stems, denoise recordings and interpolate video — all offline, all free, all from the right-click menu.
Remove Background
Cut out any subject from a photo with a single right-click. Choose between Fast mode (GPU via DirectML), High Resolution Matting for fine hair and edge detail, or High Resolution General for complex scenes. Output is a transparent PNG saved next to the source — nothing uploaded, ever.
Upscale Image
Enlarge any image ×2, ×3, ×4 or to a custom resolution using a local AI model of your choice. Real-ESRGAN x4plus for general photos and AI art, Anime 6B for illustrations and manga, Swin2SR for restoration-grade quality on degraded images. GPU-accelerated via DirectML, with automatic invisible tiling for large files.
Remove Object
Open the visual editor, paint a mask over any unwanted object with a brush, click Apply — LaMa inpainting reconstructs the background behind it. Choose Quality mode (LaMa FP32) for the best results, or Fast mode (LaMa 2025) for quicker edits. No Photoshop, no subscription, no upload.
Interpolate Video (RIFE)
Generate smoother motion and higher frame rates from any video using the RIFE model running locally via ONNX Runtime. Pick ×2, ×4 or ×8 frame rate multiplier, choose your model variant, and get the interpolated video next to the source. GPU-accelerated where available, automatic CPU fallback.
Audio Separation
Split any audio or video file into vocals, drums, bass, other and instrumental using HTDemucs ONNX exports. CPU and GPU models included. No upload, no per-track charge.
Remove Noise (Audio)
Strip background noise from any recording — room hum, fan noise, keyboard clicks, street sound. DeepFilterNet3 runs on CPU, works on any Windows machine, no GPU required.
Remove Noise (Video)
Denoise the audio track of a video file without re-encoding the video stream. The video quality is preserved exactly — only the audio is cleaned by DeepFilterNet3.
Under the hood — models and technology
| Feature | Model | License | GPU |
|---|---|---|---|
| Remove Background | BiRefNet Lite / BiRefNet HR | MIT | ✓ DirectML (fast mode) |
| Upscale Image | Real-ESRGAN x4plus | BSD-3-Clause | ✓ DirectML |
| Upscale Image | Real-ESRGAN Anime 6B | BSD-3-Clause | ✓ DirectML |
| Upscale Image | Swin2SR | Apache-2.0 | ✓ DirectML |
| Remove Object | LaMa FP32 (Quality) | Apache-2.0 | ✓ DirectML |
| Remove Object | LaMa 2025 (Fast) | Apache-2.0 | ✓ DirectML |
| Audio Separation | HTDemucs ONNX | MIT | ✓ DirectML |
| Remove Noise (Audio + Video) | DeepFilterNet3 | MIT | CPU only |
| RIFE Interpolation | RIFE v4.26 / v4.25 Lite | MIT | ✓ DirectML |
All models are downloaded on first use, stored locally, and verified with SHA-256 before each run. Models are never automatically updated without your knowledge.
Why local AI matters
Your files never leave your machine
Cloud AI tools (remove.bg, LALAL.AI, Adobe Enhance, Topaz…) upload your images, videos and audio to remote servers. FrameShift processes everything locally. Your files are never transmitted anywhere.
No subscription, no per-use charge
Cloud AI services charge monthly or per operation. FrameShift is completely free with no usage limits. Process a hundred images or a hundred songs — the cost is the same: zero.
Works offline, always available
Once a model is downloaded, every AI feature works without any internet connection. No service outages, no API rate limits, no account to log into.
Frequently asked questions
Does FrameShift send images or audio to the cloud for AI processing?
No. Every AI feature in FrameShift runs entirely on your machine. Your files never leave your disk. There is no upload, no cloud service, no remote API call of any kind.
What local AI features does FrameShift include?
FrameShift includes 7 local AI features: Remove Background (BiRefNet), Upscale Image (Real-ESRGAN, Anime 6B, Swin2SR), Remove Object (LaMa inpainting), Audio Separation (HTDemucs), Remove Noise from audio files (DeepFilterNet3), Remove Noise from video files (DeepFilterNet3), and Interpolate Video with RIFE.
Does local AI require a powerful GPU?
No. FrameShift uses ONNX Runtime with DirectML for GPU acceleration when a compatible GPU is available, and falls back to CPU automatically. All AI features work on any Windows 10/11 machine. A GPU speeds up processing significantly but is never required.
Are the AI models free to use?
Yes. All bundled models are open source with permissive licenses (BiRefNet, Real-ESRGAN, LaMa, HTDemucs, DeepFilterNet3, RIFE). They are downloaded automatically on first use and verified with SHA-256. The only exception is the optional BRIA RMBG-2.0 model, which is user-supplied and non-commercial only.
Do I need an internet connection to use AI features?
Only the first time, to download the model. After that, every AI feature runs completely offline with no internet connection required.
What technology powers the AI features in FrameShift?
All AI features use ONNX Runtime as the inference engine, with DirectML for GPU acceleration on Windows. Models are distributed as ONNX files, downloaded on first use and integrity-checked with SHA-256 before every run.
7 local AI features. Free. No upload.
Download FrameShift and run AI on your machine — no account, no cloud, no subscription.