Part I · Sovereign agent · 03
Images for content
One purpose per set. Imagine (or your still backend). Prompts saved next to files.
25 min
One purpose per set
01
Purpose
One aspect ratio
02
Style card
Locked palette
03
Batch
6–10 raw
04
Cull
3–6 winners
05
Log
prompts-used.md
A folder of random “cool AI images” is not a content system. Pick one job for this module: hero for a landing, square social, or slide background. Mixed purposes = inconsistent crop and style.
| Purpose | Aspect | Count to keep |
|---|---|---|
| Hero / wide banner | 16:9 or 21:9 | 2–3 winners |
| Social square | 1:1 | 3–4 winners |
| Slide background | 16:9, low detail behind text | 2 winners |
Style card (paste into the agent)
IMAGE STYLE CARD
Purpose: [hero | social | slide bg]
Aspect: [...]
Subject: [...]
Mood: [...]
Must include: [...]
Must avoid: text-in-image, watermarks, logos, busy faces unless requested
Consistency: same palette as deck if deck exists; lighting: [...]
Output: save under design-lab/images/ with clear names; log every prompt in prompts-used.md
Production loop
Steps · tap to check0/5
No MP4, no motion, no “animate this.” Stills only. Course 03 owns timelines.
Common failures
- Unreadable text baked into the image, ban text-in-image in the style card.
- Different style every prompt, lock palette and lighting in the card.
- Winners only in chat history, if the file is not under design-lab/images/, it does not count.
Proof of completion
design-lab/images/ has 3–6 final stills, rejects/ optional, and prompts-used.md with purpose, aspect ratio, prompt text, and which files won.
Proof · this device0/1
Stored on this device only, not sent anywhere.