A web-based tool for building, organizing, and managing prompts for image generation models like FLUX and Stable Diffusion. It treats prompts as composable structures — reusable text blocks that can be assembled, reordered, and transformed with AI assistance — rather than as flat strings you have to edit by hand. The result is a prompt workflow that scales with complexity: wildcards for randomization, snapshots for versioning, templates for reuse, and LLM-powered transforms for iteration.
The live instance is at prompts.rodeo.
- Build prompts from reusable, ordered text blocks that can be toggled on/off, reordered via drag-and-drop, and merged
- Two output styles: T5 (FLUX-optimized natural language) and CLIP (Stable Diffusion comma-separated tokens)
- Live rendered output panel with copy-to-clipboard and wildcard randomization
- Organize prompts and blocks into folders
- Define wildcard templates in JSON, YAML, line-separated, or plain text formats
- Insert wildcards into blocks with
{{wildcard:id}}syntax — resolved at render time with random selection - AI-powered wildcard generation: describe a category and get 20 values with auto-labeling
- 5 transform operations: expand, condense, and three levels of variation (slight, fair, very)
- Generation tools: explore (5 progressive variations), generate from concept, auto-label, enrich
- Works with Google Vertex AI (Gemini), OpenAI, Anthropic (Claude), Grok (xAI), and LM Studio (local)
- Users can configure their own API keys per provider, or fall back to server-configured defaults
- Text selection menu for in-place transforms on highlighted text
- Full revision history on blocks and prompts with one-click restore
- Named snapshots for capturing prompt state at a point in time
- Reusable templates created from existing prompts
- Token-based authentication (no passwords) with encrypted session storage
- Per-provider API key storage, encrypted at rest (AES-256-GCM)
- Rate limiting on auth and mutation endpoints
- Cloudflare Turnstile bot protection
- Optional client-side inference via Transformers.js for Hugging Face models
- No server round-trip needed for supported operations
Prompt Manager integrates with ComfyUI through a custom node package that pulls your active prompt directly into ComfyUI workflows. The integration uses API key authentication with real-time updates via Server-Sent Events, so changes in Prompt Manager are immediately available in ComfyUI. See the custom node repo for installation and usage details.
- Node.js 24+
- Docker (for PostgreSQL and Redis)
- Install dependencies:
npm install- Copy the example environment file and fill in the required secrets:
cp .env.example .envYou'll need to set at minimum:
SESSION_SECRET— random string, 32+ charactersENCRYPTION_SALT— exactly 32 charactersTOKEN_SECRET— random string, 32+ charactersAPI_KEY_SECRET— exactly 32 characters
LLM provider keys are optional — the app works without them, but LLM features will be unavailable.
- Start PostgreSQL and Redis:
cd local-dev && docker compose up -d && cd ..This starts PostgreSQL 16 on port 5432 and Redis 7 on port 6379. The database is created automatically.
- Start the backend:
npm run serverThe API server runs at http://localhost:3001. Database migrations run automatically on first start.
- Start the frontend (in a separate terminal):
npm run devThe app will be available at http://localhost:5173.
| Command | Description |
|---|---|
npm run build |
Build frontend and server for production |
npm run lint |
Run ESLint |
npm run tslint |
TypeScript type checking (frontend + server) |
npm run test:llm |
Run LLM service tests |
npm run migrate |
Run pending database migrations |
npm run migrate:down |
Roll back last migration |
npm run migrate:check |
Check migration status |