imagen

README

Google Imagen Provider

Google Imagen 4.0 provider for image generation using the Gemini API.

Requirements

✅ Works with AI Studio API keys - No Google Cloud billing required!

• Google AI Studio API key (same as Gemini)
• Access via generativelanguage.googleapis.com endpoint
• Model: imagen-4.0-generate-001

Configuration

apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: Provider
metadata:
  name: imagen-provider
spec:
  type: imagen
  model: imagen-4.0-generate-001

Note: Project ID and location are no longer required when using the Gemini API endpoint.

Environment Variables

• GOOGLE_API_KEY or GEMINI_API_KEY - Required for authentication

Cost

• $0.04 per image (fixed cost, not token-based)
• No token counting

API Details

• Endpoint: https://generativelanguage.googleapis.com/v1beta/models/{model}:predict
• Authentication: x-goog-api-key header (same as Gemini)
• Method: POST
• Request format: {"instances": [{"prompt": "..."}], "parameters": {"sampleCount": 1, ...}}
• Response format: {"predictions": [{"bytesBase64Encoded": "...", "mimeType": "image/png"}]}

Limitations

• No streaming support
• Works with AI Studio API keys (no GCP billing required!)
• Only supports image generation (no analysis or multimodal understanding)
• Always generates PNG format - cannot generate JPEG, WebP, or other formats
• Cannot convert or manipulate existing images
• Generates one image per request (even if prompt asks for multiple)

Alternative for Testing

For testing image generation scenarios without GCP billing:

• Use the mock provider with pre-generated images
• Or implement an OpenAI DALL-E provider (requires OpenAI API key)

Documentation

Overview

Package imagen provides Google Imagen image generation provider integration.

Index

• type Config
• type Provider
  • func NewProvider(config Config) *Provider
  • func (p *Provider) CalculateCost(inputTokens, outputTokens, cachedTokens int) types.CostInfo
  • func (p *Provider) Close() error
  • func (p *Provider) Predict(ctx context.Context, req providers.PredictionRequest) (providers.PredictionResponse, error)
  • func (p *Provider) PredictStream(ctx context.Context, req providers.PredictionRequest) (<-chan providers.StreamChunk, error)
  • func (p *Provider) SupportsStreaming() bool

Types

type Config

type Config struct {
	ID               string
	Model            string
	BaseURL          string
	ApiKey           string
	ProjectID        string
	Location         string
	IncludeRawOutput bool
	Defaults         providers.ProviderDefaults
}

ImagenConfig holds configuration for creating an Imagen provider

type Provider

type Provider struct {
	providers.BaseProvider
	Model      string
	BaseURL    string
	ApiKey     string
	ProjectID  string
	Location   string
	Defaults   providers.ProviderDefaults
	HTTPClient *http.Client
}

Provider implements the Provider interface for Google's Imagen image generation

func NewProvider

func NewProvider(config Config) *Provider

NewProvider creates a new Imagen provider

func (*Provider) CalculateCost

func (p *Provider) CalculateCost(inputTokens, outputTokens, cachedTokens int) types.CostInfo

CalculateCost calculates cost breakdown (simplified for Imagen)

func (*Provider) Close

func (p *Provider) Close() error

Close cleans up resources

func (*Provider) Predict

func (p *Provider) Predict(ctx context.Context, req providers.PredictionRequest) (providers.PredictionResponse, error)

Predict generates images based on the last user message

func (*Provider) PredictStream

func (p *Provider) PredictStream(
	ctx context.Context,
	req providers.PredictionRequest,
) (<-chan providers.StreamChunk, error)

PredictStream is not supported for image generation

func (*Provider) SupportsStreaming

func (p *Provider) SupportsStreaming() bool

SupportsStreaming returns false for Imagen