SmartGen

SmartGen for Image Generation (Experimental)

Experimental Feature SmartGen is currently in beta. API endpoints and behaviors may evolve as improvements are introduced.

---

Overview SmartGen offers a high level interface for generating images with advanced prompt engineering and precise control over image dimensions and style. It supports both FLUX and Stable Diffusion models.

---

Basic Example

tsCopy codeimport Cinna from 'cinna';

const cinna = new Cinna({
  apiKey: process.env['CINNA_API_KEY']
});

const response = await cinna.smartgen.generateImage({
  description: "A futuristic cyberpunk portrait of a young woman",
  image_model: "FLUX.1-dev",
  stylization_level: 3,
  detail_level: 4,
  color_level: 5,
  lighting_level: 2
});

---

Parameters and Controls SmartGen accepts the following input parameters:

tsCopy code{
  // Core settings
  description: string,   // Main description of the image
  width?: number,
  height?: number,

  // AI model options
  image_model?: string,  // Default: FLUX-1-dev
  is_sd?: boolean,       // Set to true to use Stable Diffusion
  language_model?: string,  // Default: nvidia/llama-3.1-nemotron-70b-instruct

  // Visual controls (scale from 1 to 5)
  stylization_level: number,  // Artistic style versus realism
  detail_level: number,       // Level of intricacy in visual details
  color_level: number,        // Color intensity from grayscale to vivid
  lighting_level: number,     // Lighting depth from neutral to dramatic

  // Optional parameters
  must_include?: string,      // Required elements in the final output
  examples?: string[],        // Reference prompts
  quality?: 'normal' | 'high',
  num_iterations?: number,    // Overrides quality if provided
  guidance_scale?: number,    // Use 3 for FLUX, 6 for Stable Diffusion
  negative_prompt?: string,   // Excludes undesired elements (only for Stable Diffusion)
  param_only?: boolean        // If true, returns parameters without generating an image
}

---

Two Step Generation

You can decouple image creation into two steps for greater control:

tsCopy code// Step 1: Generate parameters
const params = await cinna.smartgen.generateImage({
  description: "A cyberpunk cityscape",
  image_model: "FLUX-1-dev",
  stylization_level: 4,
  must_include: "neon lights, flying cars",
  param_only: true
});

// Step 2: Generate the image using the parameters
const imageResult = await cinna.images.generate({
  ...params.parameters
});

Response Format (if param_only is true)

jsonCopy code{
  "parameters": {
    "prompt": "enhanced prompt string",
    "model": "FLUX-1-dev",
    "width": 1024,
    "height": 768,
    "num_iterations": 20,
    "guidance_scale": 3,
    "neg_prompt": "optional negative prompt"
  }
}

---

One Step Generation

For direct image creation:

tsCopy codeconst result = await cinna.smartgen.generateImage({
  description: "A cyberpunk cityscape",
  image_model: "FLUX-1-dev",
  stylization_level: 4,
  must_include: "neon lights, flying cars"
});

console.log(result.url);
console.log(result.parameters);

---

Demonstration The following example illustrates the change in color intensity from monochrome to highly vibrant using color_level from 1 to 5:

Example 1: “A futuristic cyberpunk portrait of a young woman”

Example 2: “Hot air balloons in the sky”.