Local Image Studio

English | 日本語

Local Image Studio is a native macOS image-generation workspace built for Apple Silicon. A SwiftUI front end manages a local Python inference service powered by MFLUX and Apple's MLX. Generation, prompt enhancement, metadata, and image storage stay on the Mac; no cloud image-generation service is required.

Developer preview: the app is arm64-only, ad-hoc signed, and not notarized. It is intended for developers who are comfortable configuring local Python environments and model caches.

Highlights

• Local generation with FLUX.2 Klein 4B, FLUX.2 Klein 9B, and Krea 2 Turbo through MFLUX/MLX
• Generation, FLUX reference editing, variations, regeneration, export, and reusable LoRAs
• SeedVR2 7B FP16 upscaling with 2× and 4× output options
• Project-based image library with history, generation lineage, metadata, archive, and restore workflows
• Optional local prompt enhancement using models served by LM Studio or oMLX
• English and Japanese interface localization

Architecture

SwiftUI application
       │ authenticated HTTP on 127.0.0.1
       ▼
local Python service (backend_v2.py)
       │ private stdin/stdout protocol
       ▼
persistent MFLUX worker → MLX → Apple Silicon GPU

The backend listens only on the loopback interface and requires a random token generated for each launch. Generation workers enable Hugging Face and Transformers offline modes. LM Studio (127.0.0.1:1234) and oMLX (127.0.0.1:8000) are contacted only when the optional local prompt-enhancement feature is used.

Requirements

• Apple Silicon Mac running macOS 13 or later
• Apple Command Line Tools or Xcode with a mutually compatible Swift compiler and macOS SDK
• Python 3.10 or later
• MFLUX 0.19.1 with MLX 0.32.2; the default executable is ~/.local/share/uv/tools/mflux/bin/python
• Locally cached weights for each generation model you choose to use

Set LIS_MFLUX_PYTHON before launching if the MFLUX Python executable is elsewhere. Set LIS_SWIFTC when the compatible Swift compiler is not /usr/bin/swiftc.

Model weights are not bundled. Generation does not automatically download FLUX or Krea weights. SeedVR2 is the one exception: its Models screen offers an explicit Install SeedVR2 7B (~14 GB) action, which downloads the checkpoint only after the user chooses it.

Build from a clean clone

git clone https://github.com/Xombie2000/local-Image-studio.git
cd local-Image-studio
./build_v2_app.sh --build-only

The build-only command creates build/Local Image Studio.app without changing ~/Applications. To build and install the preview locally, run:

./build_v2_app.sh

The install command preserves an existing app as ~/Applications/Local Image Studio v1.app on the first replacement, then installs ~/Applications/Local Image Studio.app. Because this preview is not Developer ID signed or notarized, macOS may require approval in System Settings on first launch.

Run tests

python3 -m venv .venv
.venv/bin/python -m pip install -r requirements-dev.txt
.venv/bin/python -m unittest discover -s tests -p 'test_*.py'
./tests/run_swift_state_tests.sh

The ordinary unit tests use temporary storage and fake workers, so they do not load or download model data. The inference and telemetry scripts under tests/ are opt-in integration checks and require the configured MFLUX environment and local model caches.

Data and privacy

• Generated images: ~/Pictures/Local Image Studio/
• Metadata: ~/Library/Application Support/Local Image Studio/history.sqlite3
• LoRAs: ~/Library/Application Support/Local Image Studio/LoRAs/
• Model weights: the shared local Hugging Face cache

Active projects are .lisproject packages. Archiving stores lossless WebP images inside a ZIP container; restoring recreates PNG files at their recorded paths. The repository contains no model weights, generated-image library, or user database.

Repository layout

• macos/ — SwiftUI application, models, store, localization, and property-list resources
• backend_v2.py — local API, SQLite history, projects, jobs, and prompt-helper routing
• mflux_worker.py — persistent MFLUX generation and SeedVR2 worker
• build_v2_app.sh — command-line build, bundle, icon, signing, and optional install flow
• tests/ — Python regression tests, Swift state tests, and opt-in local inference checks
• docs/ — implementation notes, compatibility patches, and verification evidence

Model and dependency licenses

The source in this repository is MIT licensed, but that does not grant rights to third-party model weights. In particular, FLUX.2 Klein 9B uses the FLUX Non-Commercial License, and Krea 2 Turbo uses the Krea 2 Community License. Review the applicable terms before downloading or using any model, especially for commercial or production work. See THIRD_PARTY_NOTICES.md for the upstream links and license summary.

Release status

The current target is a developer preview such as v2.0.0-preview.1. A general end-user release should additionally use Developer ID signing, Apple notarization, and a documented distribution/update process.

License

Local Image Studio source code is available under the MIT License. MFLUX-derived compatibility files retain their upstream MIT notice in THIRD_PARTY_NOTICES.md.