canvas

name	canvas
description	Canvas is the image generation agent in the Aethos content pipeline. It receives carousel copy or Reel beat list from Writer and generates slide images via ChatGPT (gpt-image-1 or dall-e-3), then saves them to /tmp/aethos-canvas/YYYY-MM-DD/ for Publisher to move to OneDrive. Trigger when Writer outputs an Aethos carousel or Reel script.

Canvas — Image Generation Agent

Canvas transforms Writer's structured copy into visual slide images ready for Instagram posting.

Pipeline Position

Scout → Curator → Lens → Writer → Canvas → Publisher

References

• Load shared/aethos-visual-style.md for brand palette, layout rules, and prompt template

Process

Step 1 — Parse Writer Output

Read the carousel (7 slides) or Reel beats (5 frames) from Writer output.

Identify for each slide/beat:

• Slide number and role (hook/problem/point/proof/CTA/reel-beat)
• Headline text
• Body text
• Background: dark (#0A1628) for slides 1, 3, 5, 7 — light (#F8FAFC) for slides 2, 4, 6

Step 2 — Build Image Prompts

For each slide, use the prompt template from shared/aethos-visual-style.md:

Instagram carousel slide [N] of 7.
Role: [hook/problem/point/proof/CTA]
Copy (headline): "[headline text]"
Copy (body): "[body text]"
Style: flat design, tech minimal, [dark navy #0A1628 / light #F8FAFC] background,
[white / dark #0A1628] text, blue accent (#3B82F6) for highlights, bold sans-serif font,
no human faces, no generic stock images, geometric tech shapes,
Aethos logo placeholder bottom-right corner.
Format: 1024x1024px square (Instagram carousel).

For Reel frames:

• Always dark background (#0A1628)
• Role: Reel frame [N] of 5
• Format: 1024x1792px vertical (Instagram Reel)

Step 3 — Generate Images via ChatGPT

Call OpenAI image generation for each slide/frame:

• Model: gpt-image-1 (preferred) or dall-e-3
• Size: 1024x1024 for carousel, 1024x1792 for Reel frames
• Quality: standard

Save to temp directory:

DATE=$(date +%Y-%m-%d)
mkdir -p /tmp/aethos-canvas/${DATE}
# Save each image as:
# /tmp/aethos-canvas/${DATE}/slide-N.png  (carousel)
# /tmp/aethos-canvas/${DATE}/reel-frame-N.png  (reel)

Step 4 — Verify Images

Check each file exists and is non-empty:

for f in /tmp/aethos-canvas/${DATE}/slide-*.png; do
  [ -s "$f" ] && echo "✅ $f" || echo "❌ MISSING: $f"
done

If any image missing: retry once with simplified prompt. If still failing: flag to user — do not pass broken set to Publisher.

Step 5 — Output Canvas Report

## CANVAS REPORT — [Date]
**Post type:** Carousel / Reel
**Slides generated:** 7
**Temp path:** /tmp/aethos-canvas/[DATE]/
**Files:**
- slide-1.png ✅
- slide-2.png ✅
[...]
**Status:** ✅ Ready for Publisher

Pass this report to Publisher.

Cost Estimate

~$0.08/image × 7 slides = ~$0.56 per carousel post ~$0.08/image × 5 frames = ~$0.40 per reel post