LocalStudio — Your Private, All-in-One Media Workshop

I built a desktop app called LocalStudio. It's an open-source Windows application that compresses images, converts videos, removes backgrounds with AI, upscales photos, and more — all 100% locally on your machine.

No uploads. No servers. No data leaves your PC.

The Pivot

This app started life as Image Studio — a focused tool for image optimization. But somewhere between adding video compression, format conversion, background removal, and upscaling, the name no longer fit. So I renamed it to LocalStudio. The new name reflects what it actually does: everything media-related, running entirely on your hardware.

What It Does

Images

• Compress PNG, JPG, WebP, BMP, and GIF files — batch or single, with quality control
• Convert between JPG, PNG, and WebP formats
• Upscale 2x/3x/4x using an on-device AI model (Swin2SR architecture)
• Remove backgrounds with a locally-run ONNX model (RMBG 1.4)

Videos

• Compress MP4, MOV, AVI, MKV, WebM, and more — with preset profiles and CRF tuning
• Convert between MP4, WebM, MOV, and GIF

Everything is batch-able. Select 50 images, compress them all at once. Select a dozen videos, convert them to GIF in one click. It's fast because it's local.

How It Stays Tiny

The installer is small — roughly 10 MB. That's because the heavy components are downloaded separately on first use, not bundled with the app. The upscaling models (30 MB each) are fetched the first time you upscale an image. The background removal model (50 MB) comes down when you remove your first background. And FFmpeg (~80 MB) downloads the first time you touch a video. You only download what you actually need — if all you do is compress JPEGs, you'll never touch the video stack.

Each download shows a real-time progress bar, and files are cached once retrieved. Nothing is re-downloaded unless you clear the app data.

Tech Stack

Built with Tauri v2 — the Rust-based framework that wraps a web frontend in a native OS window. The stack is straightforward:

• Desktop shell: Tauri v2 (Rust)
• Frontend: React + TypeScript + Vite
• State management: TinyBase (reactive, survives navigation)
• Database: SQLite via sqlx
• AI inference: ONNX Runtime (ort crate)
• Video processing: FFmpeg sidecar
• Image processing: Rust image crate, jpeg-encoder, png, imagequant
• HTTP downloads: reqwest with chunked streaming + 5% progress events

The frontend communicates with the Rust backend through Tauri's invoke system — 60 registered commands covering everything from database CRUD to model inference. Each command is type-safe on both sides.

All AI models run through ONNX Runtime via the ort crate. No Python, no CUDA runtime to install. The models are standard ONNX files that run on CPU (or GPU if available).

FFmpeg is managed through ffmpeg-sidecar, which downloads a static build, extracts it, and caches it in the app data directory. The download itself uses streaming HTTP with chunked reads, emitting progress events to the frontend every 5%.

Privacy As a Feature

Every file you process stays on your machine. There's no cloud backend, no telemetry endpoint, no optional "AI enhancement server." The ONNX models run locally. FFmpeg runs locally. The SQLite database sits in your app data folder.

Compared to web-based tools or even desktop apps that phone home, LocalStudio is a dead end for your data. What goes in stays in.

The Numbers

• Current version: 0.2.3
• Supported image formats: 7 (PNG, JPG, GIF, WebP, BMP, and more)
• Supported video formats: 9 (MP4, MOV, AVI, MKV, WebM, FLV, WMV, M4V, 3GP)
• Output formats: JPG / PNG / WebP for images, MP4 / WebM / MOV / GIF for video
• Total coverage: Compression, conversion, AI upscaling, AI background removal, video compression, video conversion, AI video background removal

Download

You can download the latest release from this site. Windows x64 builds are available. macOS and Linux support are on the roadmap.

If you process media locally and care about privacy, give it a shot. It's small, free, and runs on your hardware.