> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentruntime.io/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenAI Images connector

> Generate, edit, and vary images with OpenAI DALL-E and gpt-image models from AgentRuntime workflows.

The **openaiimage** MCP adapter connects to the [OpenAI Images API](https://platform.openai.com/docs/guides/images). Generate images from text prompts, edit existing images with instructions, and create variations — returning base64-encoded assets for downstream storage, email, or social publishing.

## Prerequisites

* An OpenAI account with API access and billing enabled
* An OpenAI **API key** with Images API permissions
* For edit/variation: source images as base64 PNG (DALL-E 2) or supported formats for gpt-image-1
* **project\_contributor** access in AgentRuntime

## Connect in AgentRuntime

<Steps>
  <Step title="Get an OpenAI API key">
    Create a key at [platform.openai.com](https://platform.openai.com/api-keys). Ensure your organization can use image models.
  </Step>

  <Step title="Create a connection">
    Go to **Connections** and click **New custom connection** to create an API key connection with `APIKey`. Optionally set `DefaultModel` (e.g. `gpt-image-1` or `dall-e-3`) and `APIBase`.
  </Step>

  <Step title="Add openaiimage MCP instance">
    Go to **MCP**, click **Add instance** for **openaiimage**, wire your connection on **Instance config**, set the profile **active**, and save.
  </Step>

  <Step title="Generate a test image">
    Run **mcp\_call** → `openai_text_to_image` with a short prompt to confirm billing and model access.
  </Step>
</Steps>

See [Authentication](/integrations/authentication) for binding connections to MCP instances.

## What you can build

* **Marketing assets** — LLM writes creative brief → `openai_text_to_image` → upload to Google Drive or S3.
* **Product mockups** — Reference screenshot base64 → `openai_edit_image` with style instructions.
* **A/B variants** — `openai_image_variation` on approved hero image → **human\_task** picks winner.
* **Report illustrations** — Workflow data → LLM prompt → image embedded in PDF via Browserless.

## Tools

| Tool                     | Description                                                          |
| ------------------------ | -------------------------------------------------------------------- |
| `openai_text_to_image`   | Generate images from a text prompt (gpt-image-1, DALL-E 3, DALL-E 2) |
| `openai_edit_image`      | Edit an image with a prompt; optional mask for inpainting            |
| `openai_image_variation` | Create variations of a source PNG (DALL-E 2 only)                    |

## Example

Generate a social image from an LLM-crafted prompt:

```json theme={null}
{
  "id": "openai-image",
  "type": "mcp_call",
  "name": "Generate hero image",
  "tool_name": "openai_text_to_image",
  "tool_args": {
    "prompt": "{{steps.llm-prompt.result.image_prompt}}",
    "model": "gpt-image-1",
    "size": "1024x1024",
    "output_format": "png",
    "n": 1
  },
  "depends_on": ["llm-prompt"],
  "timeout_s": 120
}
```

## Configuration

| Key            | Required | Description                           |
| -------------- | -------- | ------------------------------------- |
| `APIKey`       | Yes      | OpenAI API key                        |
| `DefaultModel` | No       | Default model when tools omit `model` |
| `APIBase`      | No       | OpenAI API base URL                   |

## Troubleshooting

| Issue                    | Fix                                                                                       |
| ------------------------ | ----------------------------------------------------------------------------------------- |
| Content policy rejection | Revise prompt; avoid disallowed content per OpenAI usage policies                         |
| Model not available      | Confirm account access to `gpt-image-1` or `dall-e-3`; fall back to `dall-e-2`            |
| Variation fails          | Variations require DALL-E 2 and a square PNG under size limits                            |
| Large payloads           | Images return as base64; store externally rather than passing through many workflow steps |

## Related

* [Integrations quickstart](/integrations/quickstart)
* [Gemini Images connector](/connectors/geminiimage) — Google image generation
* [Connector catalog](/integrations/connector-catalog)
* [Troubleshooting](/platform/troubleshooting)