Caricature Studio

A small Express 5 web app for generating caricatures from uploaded images. The server processes uploads in memory, sends the image to a provider adapter, and returns the generated image directly to the browser. It does not persist source images or generated results.

Setup

npm install
cp .env.example .env

Set the API key for the provider you want to use. The UI lets users choose between OpenAI and Google Gemini per generation request.

Optional environment variables:

• OPENAI_IMAGE_MODEL, defaults to gpt-image-1.5
• IMAGE_PROVIDER, supported values are openai and google; .env.example selects openai
• OPENAI_API_KEY, required when using OpenAI
• GOOGLE_API_KEY, required when IMAGE_PROVIDER=google
• GOOGLE_IMAGE_MODEL, defaults to gemini-2.5-flash-image
• PORT, defaults to 3000
• MAX_UPLOAD_MB, defaults to 10

Run

npm run dev

Open http://localhost:3000.

Test

npm test

The integration tests use mocked generation providers and do not call OpenAI or Google.

Styles

The supported caricature styles are:

• Ink_Wash_Classic: Expressive ink outlines with soft grayscale washes, paper texture, elegant brush strokes.
• Bold_Vector_Pop: Thick clean outlines, flat bold colors, high contrast, poster-like, minimal shading.
• Pencil_Sketch_Studio: Fine graphite lines, cross-hatching, subtle smudges, sketchbook feel.
• Charcoal_Drama: Deep blacks, smudged charcoal, heavy shadows, dramatic light.
• Watercolor_Whimsy: Loose watercolor blooms, soft edges, gentle color gradients, light paper texture.
• Woodcut_Vintage: Carved line texture, high-contrast black and white, vintage print look.
• Comic_Strip_Satire: Clean comic linework, simple cel shading, punchy expressions, speech-bubble-ready.
• Claymation_3D: Soft clay surfaces, rounded forms, subtle fingerprints, studio lighting.
• Lowpoly_Faceted: Geometric facets, angular planes, sharp edges, vivid but limited palette.
• Neon_Retro_Chrome: 80s airbrush feel, chrome highlights, neon accents, glossy finish.

Each style is exposed by the app and used as the exact style value in requests.

API

POST /api/caricatures

Multipart fields:

• image: JPEG, PNG, or WebP
• provider: openai or google
• style: one of the exact STYLE names from prompt.txt

Success response:

{
  "imageBase64": "...",
  "mimeType": "image/png"
}