OpenAI Provider

The OpenAI provider supports GPT Image models and legacy DALL-E models for image generation, editing, and variations.

Setup

import (
    "github.com/plexusone/omniimage"
    "github.com/plexusone/omniimage/provider"
)

client, err := omniimage.NewClient(omniimage.ClientConfig{
    Providers: []omniimage.ProviderConfig{
        {
            Provider: omniimage.ProviderNameOpenAI,
            APIKey:   "sk-...", // Or use OPENAI_API_KEY env var
        },
    },
})

Models

Model	Constant	Quality	Max N	Edit	Variations
GPT Image 2	ModelGPTImage2	HD	10	No	No
GPT Image 1	ModelGPTImage1	HD	10	No	No
DALL-E 3	ModelDALLE3	HD	1	No	No
DALL-E 2	ModelDALLE2	Standard	10	Yes	Yes

DALL-E 3 Retirement

DALL-E 3 was retired on March 4, 2026. Use GPT Image models for new projects.

Image Generation

Basic Generation

resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelGPTImage2,
    Prompt: "A serene Japanese garden with cherry blossoms",
    Size:   provider.Size1024x1024,
})

High Quality

resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:   omniimage.ModelGPTImage2,
    Prompt:  "Professional product photography",
    Size:    provider.Size1024x1024,
    Quality: provider.QualityHD,
})

Style Options

// Vivid - dramatic, hyper-real images
resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelDALLE3,
    Prompt: "A fantasy dragon",
    Style:  provider.StyleVivid,
})

// Natural - realistic, less dramatic
resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelDALLE3,
    Prompt: "A forest landscape",
    Style:  provider.StyleNatural,
})

Portrait and Landscape

// Portrait (vertical)
resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelGPTImage2,
    Prompt: "A full-body portrait",
    Size:   provider.Size1024x1792,
})

// Landscape (horizontal)
resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelGPTImage2,
    Prompt: "A panoramic mountain view",
    Size:   provider.Size1792x1024,
})

Image Editing

Edit existing images with masks (DALL-E 2 only).

// Edit an image - replace masked area
resp, err := client.Edit(ctx, &provider.EditRequest{
    Model:  omniimage.ModelDALLE2,
    Image:  "data:image/png;base64,...", // Base64 encoded image
    Mask:   "data:image/png;base64,...", // Transparent areas = edit region
    Prompt: "A red sports car",
    Size:   provider.Size1024x1024,
})

Multipart Upload

The Edit API traditionally requires multipart form data. For file-based uploads, you may need to use the OpenAI SDK directly or encode images as data URLs.

Image Variations

Create variations of existing images (DALL-E 2 only).

resp, err := client.Variations(ctx, &provider.VariationsRequest{
    Model: omniimage.ModelDALLE2,
    Image: "data:image/png;base64,...",
    N:     4, // Generate 4 variations
    Size:  provider.Size1024x1024,
})

Response Format

URL Response (Default)

resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:          omniimage.ModelGPTImage2,
    Prompt:         "A test image",
    ResponseFormat: provider.FormatURL, // Default
})

imageURL := resp.Images[0].URL

Base64 Response

resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:          omniimage.ModelGPTImage2,
    Prompt:         "A test image",
    ResponseFormat: provider.FormatB64JSON,
})

b64Data := resp.Images[0].B64JSON

Revised Prompts

GPT Image and DALL-E 3 may revise your prompt for safety or clarity:

resp, err := client.Generate(ctx, &provider.GenerateRequest{
    Model:  omniimage.ModelGPTImage2,
    Prompt: "A person",
})

fmt.Println("Original:", "A person")
fmt.Println("Revised:", resp.Images[0].RevisedPrompt)
// Output might be: "A person standing in a park on a sunny day..."

Supported Sizes

Size	Constant	Models
256x256	Size256x256	DALL-E 2
512x512	Size512x512	DALL-E 2
1024x1024	Size1024x1024	All
1024x1792	Size1024x1792	GPT Image, DALL-E 3
1792x1024	Size1792x1024	GPT Image, DALL-E 3

Error Handling

resp, err := client.Generate(ctx, req)
if err != nil {
    var apiErr *openai.APIError
    if errors.As(err, &apiErr) {
        switch apiErr.Code {
        case "content_policy_violation":
            log.Println("Content violates policy")
        case "rate_limit_exceeded":
            log.Println("Rate limited, retry later")
        default:
            log.Printf("API error: %s", apiErr.Message)
        }
    }
    return
}

Best Practices

1. Use GPT Image models - DALL-E 3 is deprecated
2. Be specific - Detailed prompts yield better results
3. Handle revised prompts - Check if your prompt was modified
4. Use appropriate sizes - Match size to use case
5. Implement retries - Handle rate limits gracefully