Take any AI image apart,
token by token.
Drop a reference from Flux.1, Midjourney v6 or SDXL. my-snip-box decomposes it into atomic prompt tokens with precise metadata — subject, material, camera, lighting science, down to the micro-hardware.
⌘Vpaste an image anywhere in the app to decompose it
structured top-handle handbag, on a travertine plinth, saffiano leather, three-quarter product angle, soft key light + gradient rim, gold lobster clasp, editorial studio photography
- atomic categories
- 6 atomic categories
- granularity levels
- 3 granularity levels
- local cache
- 100% local cache
Built to decode
- Flux.1
- Midjourney v6
- SDXL
- ComfyUI
- Automatic1111
- Stable Diffusion 3
Try the real decomposition flow, right in your browser.
The web demo mirrors the desktop app's layout: filter a curated gallery by tag, open a reference to see it broken into six atomic categories, click tags into a Prompt Basket, and copy a ready-to-use prompt for FLUX or SDXL. Runs entirely in your session — drop in your own image to try it live.
Open the interactive demoPrecision controls, not a black box.
Three of the things you actually touch every session. Try each one.
Compact tag syntax says the same thing in ~60% fewer tokens.
A verbose description and a compact tag string carry identical control — subject, materials, optics, hardware. Rule-based prompt assembly just drops the connective tissue a model doesn't need, so every generation costs less.
Compact tag syntax says the same thing in ~60% fewer tokens.
A verbose description and a compact tag string carry identical control — subject, materials, optics, hardware. Rule-based prompt assembly just drops the connective tissue a model doesn't need, so every generation costs less.
structured top-handle handbag, travertine plinth, saffiano leather, matte pebbled grain, gold lobster clasp, hand-painted edge coating, three-quarter product angle, 85mm macro f/8, soft key light, gradient rim, controlled specular rolloff, editorial studio photography, muted warm palette
−59% tokens for the same control — currently showing 41% of the verbose cost.
One payment. Yours forever.
A perpetual license with early-bird launch pricing. No recurring subscription, and you bring your own model key — we never mark up your usage.
Pay once, own forever
A perpetual license. Every install you make is yours to keep.
Zero recurring subscription
No monthly fee, no seat renewals, no feature paywall creep.
Bring Your Own Key — zero mark-up
Your OpenAI / Claude / Gemini key is billed by them, at their price. We add nothing.
Personal
Early birdFor individual creators reverse-engineering their own references.
Pay once · no subscription
Get the Personal license- Atomic decomposition — subject, material, camera, lighting, hardware, style
- All 3 granularity levels (Simple / Detailed / Ultra-Detailed)
- Compact, rule-based prompt assembly + one-click copy
- Multilingual Token Bridge (KO, JA, and more) with local translation cache
- Local-first SQLite cache — offline, exportable, yours to purge
- BYOK: OpenAI, Claude, Gemini — or a fully local Ollama model
- 1.x feature updates included
- Commercial / client work license
- Batch folder decomposition
- Private & self-hosted model endpoints
Pro
Early birdFor freelancers and studios shipping client work at volume.
Pay once · no subscription
Get the Pro license- Everything in Personal
- Commercial license for client & studio work
- Batch decomposition — point at a folder, export CSV / JSON
- Private & self-hosted model endpoints (Azure, vLLM, custom base URL)
- Bulk multilingual export for localization pipelines
- Priority updates + direct roadmap input
14-day, no-questions-asked refund.
Full comparison
| Feature | Personal | Pro |
|---|---|---|
| Atomic token decomposition | ||
| Granularity levels | 3 | 3 |
| Compact, rule-based tag format | ||
| Multilingual Token Bridge | ||
| Local SQLite cache | ||
| BYOK cloud models | OpenAI · Claude · Gemini | OpenAI · Claude · Gemini |
| Local Ollama models | ||
| Commercial license | ||
| Batch folder decomposition | ||
| Private / self-hosted endpoints | ||
| Bulk multilingual export | ||
| Update channel | 1.x updates | Priority |
Runs on your machine, your keys, your models.
A lightweight Tauri desktop app. Bring a cloud key or point it at a local model — nothing leaves your device except the model calls you configure.
Platforms
01Tauri-based — native webview, no bundled Chromium.
- macOS
- 12 Monterey or later · Apple Silicon + Intel
- Windows
- 10 / 11 · x64
- Installer
- ≈ 8 MB · no runtime to install
- Idle memory
- ≈ 120 MB
Models — cloud (BYOK)
02Your key, your bill, zero mark-up.
- OpenAI
- GPT-4o / o-series vision
- Anthropic
- Claude (vision-capable models)
- Gemini 1.5 / 2.x
- Custom base URL
- Azure · vLLM · proxies (Pro)
Models — local
03Fully offline decomposition path.
- Runtime
- Ollama (local endpoint)
- Recommended
- Vision models — qwen2-vl, llava, moondream
- Network
- No external calls when local model is selected
- Fallback
- Auto-switch to cloud key if configured
Storage & privacy
04Everything stays on your disk unless you export it.
- Cache
- Local SQLite · user-chosen location
- Export
- JSON · CSV · packed-string clipboard
- API keys
- OS keychain (Keychain / Credential Manager)
- Telemetry
- Off by default · no image ever uploaded
Questions, answered.
Platform support, what stays local, how licensing and keys work.
Still stuck? Reach us on Discord.
my-snip-box is a desktop app for Windows 10/11 (x64) and macOS 12+ (Apple Silicon and Intel). It's built on Tauri, so the installer is small and it stays light on memory. There is no web, mobile, or Linux build yet.
Stop guessing at prompts. Start reading them.
my-snip-box runs entirely on your machine. Download the app, connect a cloud key or a local Ollama model, and decompose your first reference in under a minute.
Free tier included · no account required to try