Openrouter Image API

OpenRouter Image Generation API

OpenRouter provides a dedicated Image API at /api/v1/images (separate from the chat completions endpoint) for text-to-image and image-to-image generation.

Base URL: https://openrouter.ai Image endpoint: POST /api/v1/images Image models list: GET /api/v1/images/models

Quick Start

const response = await fetch("https://openrouter.ai/api/v1/images", {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${OPENROUTER_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "black-forest-labs/flux-1.1-pro",
    prompt: "a red panda astronaut floating in space, studio lighting",
  }),
});

const result = await response.json();
// result.data[0].b64_json — base64-encoded PNG image

Response images are returned as base64-encoded bytes in data[0].b64_json.

Model Discovery

Dedicated Image Models Endpoint (preferred)

GET https://openrouter.ai/api/v1/images/models

Returns all image-capable models with full capability descriptors, per-endpoint pricing, and supported parameters. Always use this to check what a model supports before using it.

General Models API (filtered)

GET https://openrouter.ai/api/v1/models?output_modalities=image

Per-Endpoint Details

GET https://openrouter.ai/api/v1/images/models/{model-id}/endpoints

Returns provider-by-provider breakdown with definitive supported params, pricing, and passthrough options. Use this when you need to know exact cost or provider-specific params.

→ See references/models.md for the current model list with capabilities and pricing.

Request Parameters

Always check supported_parameters from the models endpoint — not every model supports every field. Unsupported params are silently ignored or may error.

Response Format

Standard (buffered)

{
  "created": 1748372400,
  "data": [
    { "b64_json": "<base64-encoded-image>" }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 4175,
    "total_tokens": 4175,
    "cost": 0.04
  }
}

For vector outputs (e.g. Recraft SVG models), media_type is included:

{ "b64_json": "...", "media_type": "image/svg+xml" }

Decoding in TypeScript

const imgData = result.data[0].b64_json;
// In Node.js:
const buffer = Buffer.from(imgData, "base64");
await fs.writeFile("output.png", buffer);

// In browser:
const img = document.createElement("img");
img.src = `data:image/png;base64,${imgData}`;

Image Configuration

Resolution & Aspect Ratio

{
  model: "bytedance-seed/seedream-4.5",
  prompt: "a landscape photo",
  resolution: "2K",
  aspect_ratio: "16:9"
}

• resolution and aspect_ratio can be combined
• size is a convenience alias — pass "2K" or "2048x2048". Don't mix size with resolution/aspect_ratio — conflicting values are rejected.

Quality & Format

{
  model: "openai/gpt-image-1",
  prompt: "a product photo on white background",
  quality: "high",
  output_format: "png",
  background: "transparent"
}

Multiple Images

{ model: "openai/gpt-image-1", prompt: "a cute cat", n: 4 }
// result.data is an array of n images

Image-to-Image (Reference Images)

Pass reference images via input_references:

{
  model: "openai/gpt-image-1",
  prompt: "make this scene look like a watercolor painting",
  input_references: [
    {
      type: "image_url",
      image_url: {
        url: "https://example.com/photo.jpg"
        // OR base64: "data:image/png;base64,..."
      }
    }
  ]
}

Number of accepted references varies by provider — check supported_parameters.

Streaming Image Generation

Models with supports_streaming: true return partial renders via SSE:

const response = await fetch("https://openrouter.ai/api/v1/images", {
  method: "POST",
  headers: { "Authorization": `Bearer ${KEY}`, "Content-Type": "application/json" },
  body: JSON.stringify({
    model: "openai/gpt-image-1",
    prompt: "a detailed landscape painting",
    stream: true,
  }),
});

for await (const chunk of response.body!) {
  const text = new TextDecoder().decode(chunk);
  for (const line of text.split("\n")) {
    if (!line.startsWith("data: ") || line === "data: [DONE]") continue;
    const event = JSON.parse(line.slice(6));
    if (event.type === "image_generation.partial_image") {
      // event.b64_json — partial render, update preview
    } else if (event.type === "image_generation.completed") {
      // event.b64_json — final image, event.usage.cost
    } else if (event.type === "error") {
      // event.error.message
    }
  }
}

SSE event types:

• image_generation.partial_image — progressive render chunk
• image_generation.completed — final image + usage/cost
• error — generation failed
• data: [DONE] — stream end sentinel

Provider-Specific Options

Pass provider-specific params using the provider_slug from the endpoints API:

{
  model: "black-forest-labs/flux.2-pro",
  prompt: "a dramatic portrait",
  provider: {
    options: {
      "black-forest-labs": {
        steps: 40,
        guidance: 3
      }
    }
  }
}

Check allowed_passthrough_parameters in the endpoint record for accepted keys.

Capability Descriptor Types

When reading supported_parameters from the API:

An absent key means that parameter is unsupported by that endpoint.

Pricing Notes

• Pricing is per-endpoint (provider), not per model
• billable types: output_image, input_image, input_reference
• unit types: image, megapixel, token
• Resolution-tiered models have variant tiers (e.g. 2k, 4k) in the pricing array
• usage.cost in the response is the actual USD charged

Gotchas & Common Issues

• ❌ Wrong endpoint — image generation uses /api/v1/images, NOT /api/v1/chat/completions
• ❌ Unsupported params silently vary — always check supported_parameters first
• ❌ Conflicting size params — don't use size together with resolution/aspect_ratio
• ❌ Transparent background — only works with output_format: "png" or "webp"
• ❌ n > 1 — not universally supported; check model caps
• ✅ Model IDs change — always fetch /api/v1/images/models at runtime for current list

Reference Files

• references/models.md — Current image models, providers, key capabilities, and pricing tiers
• references/examples.md — Full TypeScript/Node.js code examples for common patterns