Create Instagram Carousel Post (Topaz Enhanced)

End-to-end pipeline: itinerary URL → photo sourcing → Topaz AI enhance → text overlays → Instagram carousel publish.

Identical to itinerary-carousel-post except Sub-agent 1 includes a Topaz enhancement step after selecting the best photo for each subject.

Parameters

• - itinerary_url (required): tabiji.ai itinerary URL (e.g. https://tabiji.ai/i/thaw-dome/)
• destination (required): City/region name (e.g. "Kuala Lumpur")
• attractions (required): List of 5 attraction names + short descriptions
• caption (optional): Custom caption. If omitted, generate one with destination name, attraction list, CTA to link in bio, and relevant hashtags.

Pipeline (3 chained sub-agents recommended)

Split into 3 sub-agents for reliability. Each writes outputs to /tmp/ig-carousel/.

Sub-agent 1: Photo Finding + Topaz Enhancement

Use the instagram-photo-find skill workflow for each subject (1 destination + 5 attractions = 6 total).

For each subject:

1. 1. web_search: site:instagram.com/p/ "{subject}" photo (10 results)
2. Download top 5 candidates: INLINECODE6
3. Vision-score each with: "Rate 1-10 as hero destination photo for {subject}. Description + score only."
4. Keep best per subject → INLINECODE7

5. Topaz Enhance each best image:
CODEBLOCK0

If the sync endpoint times out or returns a process_id instead of image bytes, use the async flow:

CODEBLOCK1

Model choice: Low Resolution V2 — designed for web-sourced images (exactly our use case). Handles JPEG compression artifacts, low resolution, and general softness. Fast and cheap.

Parameters explained:

• - output_scale_factor=2 — doubles the input resolution (2x upscale). For typical IG-sourced images (~1080px), this produces ~2160px which gives the overlay step plenty of resolution to work with.
• INLINECODE11 — keeps file size reasonable for IG's 8MB limit

Output: 6 enhanced images at /tmp/ig-carousel/{slug}-enhanced.jpg + JSON manifest at /tmp/ig-carousel/manifest.json:
CODEBLOCK2

Sub-agent 2: Text Overlays

Read manifest from sub-agent 1. Run overlay script for each enhanced image.

Slide 1 (cover) — clean style:
CODEBLOCK3

Slides 2–6 — quote style per attraction with insider tip:
CODEBLOCK4

Output: 6 overlay images at INLINECODE14

Sub-agent 3: Publish to Instagram

1. 1. Host images publicly — copy slides to tabiji repo (img/instagram/), git push, use raw GitHub URLs (https://raw.githubusercontent.com/psyduckler/tabiji/main/img/instagram/slide-{N}.jpg). Wait ~30s after push for GitHub CDN.

1. 2. Create carousel item containers (one per slide):

1. 3. Create carousel container with all children + caption:

1. 4. Publish:

1. 5. Get permalink (or verify publish on rate-limit error):

If media_publish returns a POST_ID, get the permalink directly:
CODEBLOCK8

If media_publish returns error 2207051 (rate limit / action blocked): Instagram sometimes processes the request despite returning an error. Always verify by checking the account's recent media before declaring failure:

curl -s "https://graph.facebook.com/v21.0/${IG_USER}/media?fields=id,timestamp,permalink&limit=1&access_token=${IG_TOKEN}"

1. 6. Cleanup hosted images — after publish is confirmed, delete the images from the tabiji repo and push:

cd /path/to/tabiji/repo
git rm img/instagram/slide-*.jpg
git commit -m "cleanup: remove instagram carousel images after publish"
git push

Output: Instagram post URL

Instagram API Auth

Keys from macOS Keychain:

• - instagram-access-token — Graph API token
• INLINECODE21 — IG user ID (17841449394591017)

Topaz API Auth

Key from macOS Keychain:

• - topaz-api-key — Topaz Labs API key

Caption Template

CODEBLOCK12

Tips

• - Raw GitHub URLs work for IG image_url; tabiji.ai Cloudflare CDN may trigger format validation errors.
• Add sleep 1 between container creation calls to avoid rate limits.
• If a subject yields low photo scores (<5), broaden search: try Unsplash/Flickr or more specific landmark names.
• Islamic/cultural museums tend to have fewer quality IG photos — try searching the museum's official IG handle.
• Topaz sync endpoint may return image bytes directly (check Content-Type header). If it returns JSON with process_id, switch to async flow.
• Topaz rate limits: If you get HTTP 429, use exponential backoff. Processing 6 images sequentially should be fine.
• Keep originals: The manifest stores both path (enhanced) and original so you can compare quality or fall back if Topaz fails on a specific image.