English · 中文
A lightweight, API-first image generation studio. Like Threads, it lets you advance multiple ideas in parallel within a single conversation — no need to wait for one image to finish before sending the next request, referencing an old image, or spawning a variant from any previous result. Built for users who have moved past the "tweak every parameter" phase and just want to iterate on ideas quickly through conversation.
Pairs perfectly with Infinite Image Browsing: every image SpindleStudio generates carries SD WebUI–compatible EXIF metadata, so your results drop straight into IIB for blazing-fast browsing, searching and tagging.
Download: prebuilt Windows installer / portable from GitHub Releases.
macOS users: the app is not code-signed, so Gatekeeper reports it as "damaged and can't be opened". After installing, run
xattr -cr /Applications/SpindleStudio.apponce to clear the quarantine flag. Prefer the.dmgover the.zip(some unzip tools break the app's symlinks).
SpindleStudio is not a replacement for Stable Diffusion WebUI / ComfyUI, nor a do-everything AI client. It is deliberately focused:
- 🧵 Multi-idea parallelism: multiple in-flight generation requests can progress inside the same session without waiting in line
- 💬 Conversational iteration: every message is an independent thread; continue generating from, reference, or backfill any previous result
- 🖼️ Image-generation-first: text-to-image and image-to-image, done well
- 🔌 Multi-provider: configure multiple backends and switch or call them as needed
If you have used all-in-one clients like Cherry Studio, think of SpindleStudio as the image-generation-specialized alternative.
- Multi-thread generation in one session: multiple in-flight requests progress in parallel, Threads-style
- Conversational image generation — every message is a prompt you can iterate on
- Image-to-image via referencing generated images or uploading / pasting local ones
- Batch generation (1–8 images) with progressive per-image rendering
- "Generate one more" on any previous result to append a variant instantly
- In-place prompt edit & resend, backfill to input box, delete, retry
- Generation controls: auto + 13 aspect-ratio presets, 1K / 2K / 4K resolution (4K widescreen-only)
- EXIF metadata baked into every image — prompt, dimensions and model in SD WebUI–compatible format, ready for tools like Infinite Image Browsing
- Upload history sorted by usage frequency for one-click reuse of reference images
- Local disk persistence — conversations and in-flight tasks survive refresh / restart
- Multi-window real-time sync — desktop windows and the web UI side by side, everything updates everywhere instantly, per-window themes
- Four themes: ChatGPT / Frutiger Aero / Windows Vista / Windows XP
- English & 中文 UI — follow system or set manually
- In-app settings for API key & base URL
Threads-style parallel generation — fire off several prompts at once; each runs as its own task and lands whenever it's ready:
Image-to-image — hit the link button on any previous result (or upload / paste your own image) and describe the change:
 |
 |
Batch up to 8 per message, each rendering in progressively — and the ⊕ button on any result appends one more variant; hovering a message reveals backfill / edit & resend / delete:
 |
 |
Dial in the output — 13 aspect-ratio presets with 1K / 2K / 4K resolution; every image remembers how it was made — prompt, model, dimensions and more, in SD WebUI–compatible EXIF:
 |
 |
Open as many windows as you like — extra desktop windows from the sidebar (New Window), plus the same app in the browser at http://localhost:3210 — and every window stays in real-time sync. Send a prompt in one and watch the images land in all the others; new conversations, edits and deletions appear everywhere instantly. Theme is per window, so the browser can run Windows XP while the app stays ChatGPT.
Four built-in themes, switchable anytime from the top-right dropdown:
|
 |
 |
 |
Multiple backends can live side by side in Settings — each with its own API key, base URL and model list — and the active provider/model is picked per message from the input box. Both sync (OpenAI-compatible images) and async task-style APIs are supported, and custom providers can be added from the UI. Ships with gpt-image-2 as the default model.
Run from source:
yarn
yarn dev # web dev at http://localhost:5173Fill in your API key in the settings dialog on first launch.
yarn dev:electron # dev mode (Vite + backend + Electron)
yarn build # build frontend + backend
yarn electron # run as desktop window (embedded backend)
yarn dist:win # package Windows installer / portable (electron-builder)yarn build
yarn start # http://localhost:3210Priority: in-app settings (config.json) > .env.local > .env > environment variables. See .env.example.
| Variable | Description |
|---|---|
APIMART_API_KEY / APIMART_BASE_URL |
API Mart seed key / URL (editable in-app) |
OPENROUTER_API_KEY / OPENROUTER_BASE_URL |
OpenRouter seed key / URL (editable in-app) |
OPENAI_API_KEY / OPENAI_BASE_URL |
Optional, used only for the "AI title" feature |
PORT |
Backend port, default 3210 |
HTTPS_PROXY |
Optional outbound proxy |
Data (conversations, generated images, uploads, config) lives in the project directory by default; packaged builds use the OS userData dir.
A packaged build can point its data directory at any existing folder (IIB sd_webui_dir-style). Precedence: DATA_DIR env var > data-dir.txt next to the exe > data-dir.txt in userData > default. The file contains one absolute path, e.g.:
C:\Users\me\repo\SpindleStudio
After pointing it at this repo, the packaged build and dev mode share the same conversations / images / config. For headless Node mode (yarn start), use the DATA_DIR env var. For a one-time import instead of sharing: yarn migrate <sourceDir> [destDir].
Vue 3 · Vite · TypeScript · Pinia · Ant Design Vue · Express · socket.io (real-time progress) · sharp (EXIF) · Electron
MIT