Image Generation

Image Generation

Use When

• The desired final asset is an image or image batch.
• The request may include text prompts, uploaded images, previous outputs, product URLs, research handoffs, product images, persona images, banners, thumbnails, storyboard panels, or batch variants.
• The next decision is task class, model/reference policy, and runner handoff.

Do Not Use When

• The final asset is video, audio, subtitles, transcripts, or an edit plan.
• The image request is already normalized and ready to execute. Use image-batch-runner.
• The user needs hook or benchmark decoding first. Use reference-decode.

Core Boundary

This is the image generation controller. It does not submit jobs.

It must:

1. identify the input type,
2. classify the image task,
3. select model and reference rules,
4. create the handoff for image-batch-runner.

Task Classes

Model And Reference Rules

• Text-only drafts may use a text-to-image endpoint.
• Edits require a bound source image and an edit-capable endpoint.
• Persona, product, and brand identity are binding references unless the user explicitly marks them inspiration-only.
• Benchmark clips, mood boards, and competitor references are inspiration-only unless the contract says otherwise.
• Excluded references must not enter the runner request.

Routing Table

Output Shape

Return:

• taskClass
• inputType
• modelRule
• referencePolicy
• requiredRunnerInputs
• handoffSkill
• mustNotDo

Stop Conditions

• Stop when required user intent, source evidence, or owned input artifacts are missing and guessing would change the result.
• Do not let image-batch-runner make creative classification decisions.

Public Command Boundary

• Choose the smallest matching command or workflow from the user input and run it directly.

• This public skill is instruction-driven. Produce the controller handoff artifact directly from the available evidence.

• Do not call private provider/runtime paths or unpublished local tools.

• If the CLI returns a quote-confirmation challenge, obtain user approval for its scope and cost before running postplus quote confirm --json --challenge-file <challenge.json> and retry with the returned token.