Estimated credits: ~15 per run.

• Emotion-first: shocked/curious face if relevant + bold text = high CTR
• Text overlay: 3–5 words max, high-contrast (white/yellow on dark, or vice-versa)
• Contrast & saturation: thumbnails compete in a grid — they must pop

1. Build the image generation prompt:
   • Subject:

     {{subject_description}}

     if provided, otherwise design an object/scene that dramatizes the topic.
   • Mood: derives from

     {{channel_style}}

     .
   • Composition: rule-of-thirds, subject on left or right with empty space for text.
   • Style tags:

     {{channel_style}}, youtube thumbnail composition, ultra detailed, vibrant, high contrast, 16:9

     .
2. Call

   muapi image generate

   (model=gpt-image-2-text-to-image, aspect_ratio=16:9).

• Suggested overlay text: 3–5 bold words that complement the title

  {{title}}

  .
• Text placement: where on the canvas to position text (e.g. "bold yellow text, top-right third").
• Font recommendation: style suggestion (e.g. "Impact-style all-caps with black outline").

• Never put too much text in the prompt — text rendering in image models is unreliable. Guide the user on adding text in post-production (Canva, Photoshop).
• If the user already has a channel image or face photo in the session, use

  muapi image edit

  to incorporate it.
• Suggest A/B variants only if the user asks.

• This recipe is LLM-orchestrated: read each phase, gather any missing inputs from the user, then call

  muapi

  CLI commands. Use

  muapi auth configure

  first if

  MUAPI_API_KEY

  is unset.
• For model IDs without a CLI alias yet, fall back to the raw endpoint via

  curl -X POST https://api.muapi.ai/api/v1/<endpoint> -H "x-api-key: $MUAPI_API_KEY" -H 'content-type: application/json' -d '{...}'

  and poll with

  muapi predict wait <request_id>

  .
• Substitute

  {{input_name}}

  placeholders with the user's actual inputs before issuing each call.