A native desktop control room for AI image generation — runs OpenAI, Google Imagen, Stability, Replicate, Fal, and Black Forest Labs side-by-side, dumps results into folders you actually understand, and pushes them to your own GitHub repo if you want.

Install · Quickstart · Prompt syntax · Cutting a release

---

Why

Every image-gen vendor wants you in their browser tab. Their gallery, their file naming, their ephemeral history. You end up with seventeen open tabs and zero idea where the good ones are.

ImgPlayGround inverts that. Your prompts run here, the outputs land here, named the way you want, in folders you picked, optionally backed up to a git repo you control. The vendors are interchangeable. You aren't.

What it is

• Native desktop app — Tauri v2 (Rust + WebView). ~10MB installer, starts in ~200ms, no Electron tax.
• Six providers, one queue — OpenAI gpt-image-2, Google Imagen 4, Stability Core/Ultra/SD3, Replicate (Flux + others), Fal.ai (Flux variants), Black Forest Labs (Flux Pro 1.1 / Ultra). Bring your own API keys.
• Bulk mode that actually scales — paste N prompts, pick M providers, watch a live gauge fill in as each prompt × provider job lands.
• Storage you can find later — predictable paths, JSON sidecars with the prompt + model + seed, and a prompt syntax that lets you put images exactly where you want them.
• Optional GitHub sync — connect a token, pick or create a repo, every batch auto-commits and pushes. Token never touches .git/config.

Install

Fedora 38+

sudo dnf install -y webkit2gtk4.1-devel glib2-devel gtk3-devel libsoup3-devel \
                    openssl-devel libappindicator-gtk3-devel librsvg2-devel \
                    libsecret-devel gcc gcc-c++

git clone https://github.com/JaceDev/ImgPlayGround
cd ImgPlayGround
npm install
npm run app:dev

Debian / Ubuntu

sudo apt install -y libwebkit2gtk-4.1-dev libgtk-3-dev libsoup-3.0-dev \
                    libssl-dev libayatana-appindicator3-dev librsvg2-dev \
                    libsecret-1-dev build-essential

macOS / Windows

Once a release exists: download .dmg / .msi from Releases and double-click. Done.

To build from source you'll need Node 20+, Rust stable, and Xcode CLT (mac) or MSVC Build Tools (win).

Quickstart

1. Launch the app.
2. Keys → paste your provider API keys. Stored in your OS keyring (libsecret / Keychain / Credential Vault). Never on disk.
3. Settings → pick an output directory. This becomes your library root.
4. (Optional) Settings → GitHub sync → Connect. Use a classic token with the repo scope. Pick an existing repo or create a new one from inside the app.
5. Generate a single prompt, or Bulk an entire batch.
   • ⌘/Ctrl + ↵ sends.

Providers

Provider	Default model	Notes
OpenAI	gpt-image-2	Live /v1/models lookup
Google Imagen	imagen-4.0-generate-001	Via Gemini API
Stability AI	core	core / ultra / sd3
Replicate	black-forest-labs/flux-1.1-pro	Curated text-to-image collection
Fal.ai	fal-ai/flux-pro/v1.1	Flux family + Recraft + Ideogram
Black Forest Labs	flux-pro-1.1	Flux Pro / Ultra / Dev

The Model dropdown in Generate is populated dynamically per-provider — hit ↻ to refresh. Free text still wins, so you can punch in any model the provider exposes.

Storage layout

By default, every image goes here:

{output}/generated/{provider}/{YYYY-MM-DD}/{slug-of-prompt}-{shorthash}.png
{output}/generated/{provider}/{YYYY-MM-DD}/{slug-of-prompt}-{shorthash}.png.json

The sidecar JSON has the full prompt, model, size, seed, and timestamp — so you can find or re-run anything later.

Prompt syntax

Power-user trick: append shape suffixes to redirect where the file lands.

Suffix	Effect
:name.png	Save with that exact filename
:tiles/grass/grass_32x32.png	Save at {output}/tiles/grass/grass_32x32.png
;tiles/grass	Save under {output}/tiles/grass/ (auto-named)
;folder:name.png	Folder + name as separate suffixes

Examples:

mossy stone wall:tiles/walls/wall_a.png
golden retriever astronaut:dogs/space.jpg
isometric cyberpunk diner;scenes

The provider only sees the cleaned prompt — the suffix is stripped before the API call. n > 1 with a custom name auto-numbers as name-2.png, name-3.png. Path segments are sanitized (no .., no absolute paths).

Bulk inherits the same syntax — every line in your prompt list can have its own destination.

Gallery

• Contact-sheet layout — every 9th tile gets the 2×2 hero slot. Hover any tile and it glows in its provider's color.
• Filter bar matches prompt, provider, or path.
• Select mode lets you multi-pick and save anywhere with an optional subfolder — e.g. pick ~/Pictures, type sprites/grass, drop 12 tiles into ~/Pictures/sprites/grass/.

GitHub sync

Connect once via Settings. Two token options:

Token type	Scope	Can create repos?	Can push?
Classic ★ recommended	repo	✅	✅
Fine-grained	Administration: R/W + Contents: R/W on target repos	✅	✅

Create a classic token →

Tokens live in your OS keyring. The PAT is never written to .git/config — each push assembles a tokenized URL on the fly and the token is scrubbed from any error output.

Cutting a release

CI auto-builds and publishes on every v* tag — Linux (.AppImage + .rpm

• .deb), macOS Apple Silicon (.dmg), Windows (.msi).

# 1. Bump version in package.json, src-tauri/Cargo.toml,
#    src-tauri/tauri.conf.json. Commit.
# 2. Tag and push.
git tag v0.1.0
git push origin v0.1.0

Release notes are auto-generated by GitHub from PR titles since the previous tag, bucketed by label per .github/release.yml (✨ Features, 🐛 Fixes, 🎨 Design / UX, 🛠 Internals, 📚 Docs).

Linux + NVIDIA crash on launch

If the app insta-closes on Fedora / other distros with the proprietary NVIDIA driver and coredumpctl info points at libnvidia-eglcore.so, that's a known WebKitGTK + NVIDIA bug in the DMABuf renderer. v1.0.1+ ships with the workaround baked into the binary (WEBKIT_DISABLE_DMABUF_RENDERER=1 set at startup on Linux only).

If you're stuck on v1.0.0, launch with the env vars manually:

WEBKIT_DISABLE_DMABUF_RENDERER=1 WEBKIT_DISABLE_COMPOSITING_MODE=1 \
  /usr/bin/imgplayground

Auto-update

Installed builds quietly check the GitHub Releases API on launch. If a newer version exists, a banner slides in at the bottom right with Install → progress bar → Restart. Failures (no internet, no release yet) are swallowed silently — never bothers users with infrastructure noise.

One-time signing setup

The updater verifies bundles with a key pair. Run this once locally:

npm run tauri signer generate -- -w ~/.tauri/imgplayground.key

Then:

1. Copy the public key into src-tauri/tauri.conf.json → plugins.updater.pubkey (replace the REPLACE_WITH_OUTPUT_… placeholder).
2. Add two GitHub Actions secrets (Settings → Secrets and variables → Actions):
   • TAURI_SIGNING_PRIVATE_KEY — contents of ~/.tauri/imgplayground.key
   • TAURI_SIGNING_PRIVATE_KEY_PASSWORD — the password you set at generation

From then on, every v* tag produces signed bundles + a latest.json manifest attached to the release. Existing installs notice within seconds of being opened.

Don't lose the private key. Without it you can't ship updates to existing installs — they'll reject the new key's signature. Back it up to a password manager.

Repo layout

.
├── src/                    React UI
│   ├── App.tsx
│   ├── components/         Rail · BottomBar · Generate · Bulk · Gallery · Keys · Settings
│   └── lib/                types · tauri (invoke wrappers) · store (zustand)
├── src-tauri/
│   ├── src/
│   │   ├── lib.rs          Tauri command surface
│   │   ├── providers/      One adapter per provider
│   │   ├── github.rs       GitHub API client + tokenized push helper
│   │   ├── storage.rs      Path layout + suffix parsing + sidecars
│   │   ├── git.rs          init / commit / set-remote / push
│   │   ├── keys.rs         OS keyring wrapper
│   │   └── config.rs       Persistent app settings
│   ├── capabilities/       Tauri v2 capabilities
│   ├── icons/              App + bundle icons
│   └── tauri.conf.json
├── .github/
│   ├── workflows/          CI + Release
│   └── release.yml         Auto-changelog bucketing
└── .impeccable.md          Design context for any future UI work

Adding a new provider

1. src-tauri/src/providers/<name>.rs — implement generate(api_key, GenParams) -> Vec<GenOutput> and list_models().
2. Add the variant to Provider in providers/mod.rs plus the from_str arm.
3. Add the string to PROVIDERS, PROVIDER_LABELS, PROVIDER_DEFAULT_MODEL, and PROVIDER_COLORS in src/lib/types.ts.
4. Pick a hue (--p-<name>) in src/styles.css.
5. Rebuild.

The UI picks up the new provider automatically — including in the bulk matrix, the gallery glow, and the bottom-bar status dots.

Tech

Frontend	React 18 · TypeScript · Vite · Tailwind v3 · Zustand
Display fonts	Geist (UI) · Geist Mono (paths) · Fraunces (titles)
Backend	Rust · Tokio · reqwest (rustls) · keyring v3
Shell	Tauri v2
Bundles	AppImage · RPM · DEB · DMG (arm64 + x64) · MSI

Changelog

See CHANGELOG.md — versioned, follows Keep a Changelog.

License

MIT — see LICENSE.

Credits

Built with help from the ridiculous lineup of open-source maintainers behind Tauri, Vite, Tailwind, Zustand, Geist, Fraunces, the Rust ecosystem, and the six provider SDKs that didn't have to expose this much surface but did anyway.