OpenAI Image Generation

This reference graph tests provider-backed image generation using generic Subfork primitives rather than a hardcoded provider node.

Before running it, add a graph-scoped secret named OPENAI_API_KEY in Graph Settings > Execution Profile.

The only user-editable content required for a basic run is the Editable prompt node. Change its text parameter to change the generated image prompt.

The graph shape is:

Text Value -> JSON Template -> HTTP Request -> JSON Path -> Base64 Image To Media -> Media Response

JSON Template exposes a custom prompt input and renders it into the request body with ``. This graph is intended to become a candidate composite/subgraph node later: the authoring view can show the internals, while the higher-level graph can show a single OpenAI Image Generation node with prompt in and image out.

Expected outputs:

• image points at the Image media node’s media output for Image Preview panels.
• response points at Media Response for API-style graph execution.

Known v0 caveat: the response envelope redirects to the generated media URL. For base64 image data this is currently a data: URL, which is useful for local preview. Future artifact storage should replace large inline data URLs with artifact references.

The request body intentionally asks for a smaller response payload:

{
  "size": "1024x1024",
  "quality": "low",
  "output_format": "jpeg",
  "output_compression": 70
}

gpt-image-1 does not currently expose smaller fixed dimensions than 1024x1024, so compressed JPEG/WebP output is the practical v0 way to reduce the base64 JSON response size.