Image generation

Request

Target an image-capable model (bold/image auto-routes across providers) and declare the output modalities. The conversation is the interface: follow-up messages edit or iterate on the image with full context.

const res = await client.chat.completions.create({
  model: "bold/image",
  modalities: ["image", "text"],
  messages: [{ role: "user", content: "A lighthouse at dawn, watercolor" }],
  image_config: { aspect_ratio: "16:9" },   // optional
});

const image = res.choices[0].message.images?.[0];
// image.image_url.url → "data:image/png;base64,...."

image_config is optional: aspect_ratio (e.g. "1:1", "16:9", "9:16") is mapped to each provider's native control; size (e.g. "1024x1024") passes through where supported.

Response

Generated images arrive as base64 data URLs in the assistant message's images array - the gateway hosts nothing and stores nothing.

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "Here is your lighthouse at dawn.",
      "images": [{
        "type": "image_url",
        "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." }
      }]
    },
    "finish_reason": "stop"
  }]
}

With stream: true, image fragments ride in delta.images on a chunk, alongside the usual delta.content text tokens.

Model discovery

Filter the catalog by output modality, OpenRouter-style:

curl "https://boldrouter.com/v1/models?output_modalities=image" \
  -H "Authorization: Bearer $BOLD_API_KEY"

• bold/image - auto-routes across providers' image models with health-based failover.
• google/gemini-2.5-flash-image - conversational generation and editing, text alongside images.
• openai/gpt-image-1 - OpenAI's image model; your last user message is the prompt.

Pricing

Images bill per generated image (the image_per_unit rate in /v1/models), on top of the request's token cost. Usage events, spend caps, balance reservations, and the dashboard ledger account for image cost exactly like token cost.