Continuum - Consistent Prompt Engine

Web tool for consistent AI image generation. Define reusable assets (characters, locations, objects) with LLM-assisted prompt enrichment, create variants, and compose scenes. Ensures visual consistency across image series through structured prompt building blocks and template-based scene assembly.

Experimental - This project is in early development. Currently only tested with:

• LLM: Claude Sonnet 4.5 (via OpenRouter)
• Image Generator: Nano Banana Pro

Other models may work but are untested.

# Pull from GitHub Container Registry
docker pull ghcr.io/digijoe79/continuum:latest

# Run container
docker run -d -p 80:80 -v continuum-data:/app/data ghcr.io/digijoe79/continuum:latest

# Or build locally
git clone https://github.com/digijoe79/continuum.git
cd continuum
docker-compose up --build

Access at http://localhost

Backend:

cd backend
pip install -r requirements.txt
PYTHONPATH=. uvicorn main:app --reload --port 8000

Frontend:

cd frontend
npm install
npm run dev

Access at http://localhost:5173

Assets are reusable prompt building blocks. Each asset has a base prompt that describes its core visual characteristics.

Variants are modifications of an asset. They describe what changes, not the entire asset again.

Example:

• Base Asset: "Anna" (young woman, red hair, freckles)
• Variant "Medieval": "burgundy velvet dress, golden embroidery, leather belt"
• Variant "Modern": "white t-shirt, blue jeans, sneakers"

Scenes compose assets into final prompts using a simple template syntax:

[ANNA:Medieval] walks through [MARKET:Night]

This resolves to a complete prompt combining:

• Anna's base description
• Medieval variant details
• Market's base description
• Night variant details
• Selected shot type, lighting, and style

Pre-configured settings available across all projects:

• Shot Types: Camera framing (Close-up, Wide, POV, Dutch Angle, etc.)
• Lighting Setups: Light configuration (Golden Hour, Noir, Silhouette, etc.)
• Styles: Visual aesthetics (Cinematic, Anime, Photorealistic, etc.)

1. Create Project
   └── Define your story/series scope

2. Create Assets (Upstream - with LLM)
   ├── Add Character/Location/Object
   ├── Describe in chat dialog
   └── LLM generates structured prompt

3. Create Variants
   ├── Add variant to asset
   ├── Describe the modification
   └── LLM generates delta prompt

4. Compose Scenes (Downstream - no LLM)
   ├── Write action text with [ASSET:Variant] syntax
   ├── Select shot type, lighting, style
   └── Generate final prompt

Go to Settings to configure your LLM provider:

Use the Test Connection button to verify your configuration.

MIT