image-model-routing

Image Model Routing

Use this skill when deciding which image provider, model, surface profile, and quality tier should be used.

Inputs

• brief: creative brief or user request
• surface: optional platform/profile target
• objective: business or creative goal, such as CTR, typography, brand photo, blog hero, or social carousel
• provider: optional forced provider (auto, openai, google, or a configured runtime provider id)
• model: optional forced model
• quality: optional quality tier (draft, standard, or premium)

Procedure

1. Extract the intended platform, aspect, usage constraints, visible text needs, brand constraints, and required output quality.
2. Call bakin_exec_images_recommend with the extracted surface, objective, provider, model, and quality.
3. Use the returned provider, model, surface, width, height, and quality exactly unless the user explicitly changes the constraints.
4. If the recommendation says credentials are missing, report the missing provider route and do not invent a raw provider call.
5. Prefer runtime routes when configured. With OpenClaw, this includes Codex OAuth-backed OpenAI routes and any image provider OpenClaw reports as configured.

Output

Return a compact route object:

{
  "provider": "openai",
  "model": "gpt-image-2",
  "surface": "instagram-feed-portrait",
  "width": 1080,
  "height": 1350,
  "quality": "standard",
  "reason": "Short rationale"
}