Photonix

Shape light. Repair detail. Direct the image.

A Windows-first AI photo editor built around gpt-image-2, designed for photographers who want fast, prompt-driven edits on large landscape PNG/JPG photos with predictable, local-first workflows.

Photonix is a desktop application — your originals stay on disk, your API key stays on your machine, and edits go directly from the desktop to the provider you configure.

---

Highlights

• BYOK provider config — supply your own baseURL and API key for any OpenAI-compatible image model
• Mask-based local edits — paint a region, write a prompt, get a result that respects what you asked to preserve
• Prompt compiler — your free-text prompt is rewritten into a structured edit instruction with explicit preservation and negative constraints
• Image generation mode — separate text-to-image tab with its own gallery, persisted across restarts
• Upload proxy profiles — fast / recommended / high_quality so a 36MB landscape PNG never gets uploaded raw
• Social export presets — WeChat Moments / High Quality Mobile / Small File / Archive PNG, plus a custom mode
• Built-in landscape and portrait presets — 8 landscape + 4 portrait one-click starting points, plus your own custom presets
• Persistent prompt history — recent prompts stay around across sessions
• Small batch queue — multi-select in Library, run one preset over several photos with retry/cancel per item
• Version history — every accepted edit becomes a version; switch between draft/final/stitched/original; original files are never modified
• Local-first storage — SQLite for metadata; disk for thumbnails, proxies, masks, generations, and edited versions; secrets in a separate file outside the database

---

Screenshots

Coming soon. Add your own captures into docs/screenshots/ and reference them here.

Library	Editor

Generate	Settings

---

Tech Stack

Layer	Stack
Desktop shell	Tauri 2
Frontend	React 19 + TypeScript + Vite + Tailwind CSS 4
State	Zustand
Backend	Rust
Database	SQLite (via rusqlite, bundled)
HTTP	reqwest (Rust)
Image processing	image crate
File scanning	walkdir

Provider integration:

• POST /images/edits for prompt+mask edits
• POST /images/generations for text-to-image
• POST /chat/completions for prompt compilation
• All HTTP calls happen in Rust to avoid CORS and keep the API key out of the JS layer

---

Repository Layout

photonix-app/                 Desktop application (Tauri + React + Rust)
  src/                          Frontend
    components/{layout,library,editor,settings,generate,ui}/
    services/{tauri,prompt,provider}/
    stores/                     Zustand stores
  src-tauri/                    Rust backend
    src/commands/{edit,prompt,generate,library,secrets}.rs
    src/image_core/             Scan + thumbnail/proxy generation
    src/storage/                SQLite + migrations + repository

gpt-image-editor-prd-design.md   Full product + technical design (MVP1 + MVP2 + §31 Generate)
information-architecture-wireframes.md
storage-and-database-design.md
mvp-development-breakdown.md
schema.sql                       Initial database schema reference
BRAND.md                         Naming + tone direction (legacy: "LumaForge")

---

Getting Started

Prerequisites

• Windows 10/11 (primary target)
• Node.js 20+ (developed against 25)
• Rust stable + Visual Studio Build Tools (C++ workload + Windows SDK)
• An API key for an OpenAI-compatible provider that supports gpt-image-2 style image edit/generation

Run in development

cd photonix-app
npm install
npm run tauri dev

The first build pulls all Rust crates and may take 3–5 minutes. Subsequent builds are incremental.

Build a release installer

cd photonix-app
npm run tauri build

Output is in photonix-app/src-tauri/target/release/bundle/.

First-run configuration

1. Open the Settings screen (gear icon in the left sidebar)
2. Provider tab: enter your baseURL, API key, and confirm the image/text models
3. Click Save Settings, then Validate Connection
4. Editing tab: choose an upload proxy profile (default: Recommended)
5. Export tab: pick a default export preset (default: WeChat Moments)

API keys are stored in an obfuscated file outside the SQLite database. Production-grade Windows Credential Manager integration is part of the §32.6.9 hardening pass before public beta.

---

Workflows

Edit an existing photo

1. Library → Import a folder of photos (PNG/JPG/WebP/TIFF/BMP)
2. Wait for thumbnails (background two-pass downscale; cached on disk)
3. Double-click an image → Editor
4. Mask tab: paint the region you want edited (B brush / E erase / [ ] adjust size)
5. Prompt tab: type a prompt or pick a Landscape/Portrait preset; toggle preserve identity / preserve composition
6. Generate Draft (Ctrl/Cmd + Enter) — Photonix uploads a compressed proxy, not the original
7. The new version becomes active; History tab lets you switch between versions
8. Export tab: pick a preset, choose a save location

Generate from scratch

1. Sidebar → Generate (sparkle icon)
2. Type a prompt, pick size and quality, click Generate (or Ctrl/Cmd + Enter)
3. Result lands in the gallery and the preview area
4. Export PNG via the toolbar above the preview

Batch edit

1. Library → click Select to enter multi-select mode
2. Pick the images you want to edit
3. Batch Edit (N) → choose a preset or write a prompt
4. Start — items run sequentially with status (queued / running / succeeded / failed / canceled)
5. Failed items can be retried individually; queued items can be canceled in bulk

---

Status

MVP1 — shipped: library, mask, prompt compile, edit pipeline, version history, export, settings, BYOK, Generate tab.

MVP2 — shipped: upload proxy profiles, social export presets, 8 + 4 built-in edit presets, persistent prompt history, custom presets, small batch queue, Rust-side compatibility check.

Coming next (M6 hardening):

• Windows Credential Manager / keyring for API key storage
• Improved error logging and structured diagnostics
• Batch concurrency tuning
• Real-image tuning of the default proxy and export presets

See the design document for the full scope and roadmap:

• Full PRD + Technical Design — covers product goals, MVP1/MVP2 milestones, prompt compiler, large-image strategy, image generation (§31), and MVP2 social landscape edition (§32)
• Information Architecture & Wireframes
• Storage & Database Design

---

Privacy

• Original files are never modified.
• Edited results, generations, thumbnails, and proxies live in your local app data directory.
• Prompts and metadata live in a local SQLite database.
• The API key lives in a separate obfuscated file outside the database.
• Image content is sent only to the baseURL you configure, only when you trigger an edit or generation.

---

License

Released under the MIT License.