Image generation recipes

An image generation recipe is a specialized image feature that records the instructions and context used to create an AI-generated image. The shared GenerationRecipe schema gives extensions a common structure for capturing generation provenance across different services and applications.

The role of image recipes

An AI-generated image is the result of a process, not just a prompt. The model, prompt or workflow, generating software, parameters, and any input assets can all contribute to the final result. A recipe brings the available parts of this context together with the image.

Recording a recipe can help creators:

• understand and, where possible, reproduce how an image was generated;
• identify the model, software, and source assets involved;
• trace relationships between generated images and their inputs or derivatives;
• review provenance and support accountability about the process used.

Together, these details support tracing image origins. If an image is later transformed or used to generate another image, recording its input assets and generation context can contribute to a revision graph — a git-like history of the image and its derivatives. This history can help creators account for the steps they took and provide useful provenance when reviewing licensing, copyright-sensitive inputs, or applicable legislation.

Recipes are only as complete as the information available from the generation service or application and the extension that captures it. They document provenance; they do not establish copyright ownership, licensing status, or legal compliance.

Anatomy of an image recipe

A recipe has required fields and optional context:

Field	Required	Purpose
schemaVersion	Yes	Identifies the version of the recipe schema.
modelTags	Yes	Lists the model or model versions used for generation.
prompt	Yes	Holds the prompt or structured generation instructions.
id	No	Identifies the recipe or generation instance.
url	No	Links to the generation request or result in an external service.
software	No	Identifies the application or service that performed the generation.
author	No	Records the recipe author when known.
inceptionDate	No	Records when the recipe was created.
inputAssets	No	Lists identifiers or URLs for assets used as inputs.
aspectRatio	No	Records the output image aspect ratio.

The prompt field has two forms, identified by its kind value:

Form	Structure	Purpose
textual	A text string.	Represents a direct text prompt used by services that accept a natural-language prompt.
instructions	A structured value object.	Preserves a workflow, provider request, or tool-specific generation instructions.

The textual form provides a simple representation for prompt-based services. The instructions form can retain richer, structured inputs without reducing every generation workflow to one text string. Both forms fit the same recipe schema, allowing different extensions to record provenance in a consistent shape while preserving generation-specific details.

Textual prompt example

{
  "schemaVersion": 2,
  "modelTags": ["example/model"],
  "prompt": {
    "kind": "textual",
    "text": "A lighthouse on a rocky coast at dusk"
  }
}

Instructions prompt example

{
  "schemaVersion": 2,
  "modelTags": ["example/model"],
  "software": "ComfyUI",
  "prompt": {
    "kind": "instructions",
    "value": {
      "positivePrompt": "A lighthouse on a rocky coast at dusk",
      "negativePrompt": "low resolution",
      "seed": 12345
    }
  }
}

The contents of an instructions value depend on the generating software or service. Picteus records it as structured data; the corresponding extension or integration is responsible for understanding its specific format.

For the general feature model and validation rules, see Features. For the API endpoint that returns parsed recipes associated with an image, see the image web services reference.

Recipe storage and validation

In the feature model, a recipe is an ImageFeatureType.RECIPE feature with the json format. It is stored like other image features, but the setFeatures and ensureFeatures web services validate its value against the recipe schema before accepting it. See the feature write endpoints, the OpenAPI specification, and the recipe JSON schema for the endpoint behavior and contract.