What does referenced_image_ids mean in a ChatGPT image prompt?

By PicGens editorial · Published

Checked against the official OpenAI image-generation guide on October 12, 2026 and field names observed in shared PicGens prompt text. Undocumented fields have no verified public contract.

Sources: OpenAI image-generation guide

referenced_image_ids appears in some shared image-generation payloads. The official image-generation guide reviewed for this article does not document it as a public request parameter. Treat it as source-specific trace data: copy the creative instructions, provide your own reference image, and use the documented input format of your chosen product.

Where these field names appear

Some gallery sources share a structured request or tool trace instead of a plain creative brief. A field name in that text tells you what the creator shared; it does not establish a supported API. PicGens preserves the original text for context and copying, while displaying readable topic descriptions separately.

  • referenced_image_ids: the name suggests references to images, but its value format and lookup behavior are not established by the public guide.
  • gen_id and parent_gen_id: identifier-like fields seen in shared traces. Their relationship and persistence are not verified here.
  • image_gen.text2im: a tool-like name in a trace, rather than a documented endpoint in the guide reviewed.
  • is_style_transfer: a flag-like field whose effect is not documented in that guide. Do not assume changing it enables a public feature.

Use a reference image through a supported input

In an image-capable chat interface, attach an image and describe what to change and preserve. A copied identifier is not a substitute for supplying the image. For API work, the official guide describes Image API edits and Responses API image inputs. Responses inputs use input_image with image_url or file_id; multi-turn editing can carry image context or previous_response_id. These are documented mechanisms, not a conversion table for internal trace fields.

  1. Keep the creative brief. Extract the subject, edit, lighting and composition instructions from the source text.
  2. Supply your own input. Attach the reference image through the interface or input schema you are actually using.
  3. Review the result. Check the protected details and request one concrete correction if needed.

Why a copied identifier may not work

A trace can depend on context that the shared post does not include. Do not guess an identifier’s storage location, try to reconstruct another creator’s hidden inputs, or treat a failed lookup as proof that a model cannot edit images. Ask for the reference material or start from your own image and a plain-language brief. Model labels in gallery examples come from their sources; check the supported model and schema in the tool you use.

Frequently asked questions

Can I send referenced_image_ids to the public OpenAI API?
The official image-generation guide reviewed here does not document that parameter. Use its supported image inputs and endpoint schema instead.
Is gen_id the same as a file_id or response ID?
No equivalence has been established by the reviewed documentation. Keep these identifiers separate and use only the type expected by your documented workflow.
Should I copy the entire payload into ChatGPT?
Start with the creative instructions and attach the reference image. Internal field names are unnecessary unless your actual tool explicitly documents them.

Keep exploring