itinerary-carousel-post-topaz（行程旋转木马柱黄玉）

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. web_search: site:instagram.com/p/ "{subject}" photo (10 results)
2. Download top 5 candidates: curl -s -L -o /tmp/ig-carousel/raw-{slug}-{n}.jpg "https://www.instagram.com/p/{shortcode}/media/?size=l"
3. Vision-score each with: "Rate 1-10 as hero destination photo for {subject}. Description + score only."
4. Keep best per subject → /tmp/ig-carousel/{slug}-best.jpg

5. Topaz Enhance each best image:

TOPAZ_API_KEY=$(security find-generic-password -s "topaz-api-key" -w)

curl --request POST \
  --url https://api.topazlabs.com/image/v1/enhance \
  --header "X-API-Key: ${TOPAZ_API_KEY}" \
  --header 'accept: image/jpeg' \
  --header 'content-type: multipart/form-data' \
  --form 'model=Low Resolution V2' \
  --form 'output_scale_factor=2' \
  --form 'output_format=jpeg' \
  --form "image=@/tmp/ig-carousel/${slug}-best.jpg" \
  --output "/tmp/ig-carousel/${slug}-enhanced.jpg"

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

# Async: submit
RESPONSE=$(curl -s --request POST \
  --url https://api.topazlabs.com/image/v1/enhance/async \
  --header "X-API-Key: ${TOPAZ_API_KEY}" \
  --header 'content-type: multipart/form-data' \
  --form 'model=Low Resolution V2' \
  --form 'output_scale_factor=2' \
  --form 'output_format=jpeg' \
  --form "image=@/tmp/ig-carousel/${slug}-best.jpg")

PROCESS_ID=$(echo "$RESPONSE" | jq -r '.process_id')

# Poll status until Completed
while true; do
  STATUS=$(curl -s --header "X-API-Key: ${TOPAZ_API_KEY}" \
    "https://api.topazlabs.com/image/v1/status/${PROCESS_ID}" | jq -r '.status')
  [ "$STATUS" = "Completed" ] && break
  sleep 3
done

# Download result
curl -s --header "X-API-Key: ${TOPAZ_API_KEY}" \
  "https://api.topazlabs.com/image/v1/download/${PROCESS_ID}" \
  --output "/tmp/ig-carousel/${slug}-enhanced.jpg"

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.
• output_format=jpeg — 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:

[{"slug": "kuala-lumpur", "subject": "Kuala Lumpur", "score": 7, "path": "/tmp/ig-carousel/kuala-lumpur-enhanced.jpg", "original": "/tmp/ig-carousel/kuala-lumpur-best.jpg", "source": "instagram.com/p/XXX/", "topaz_enhanced": true}]

Sub-agent 2: Text Overlays

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

Slide 1 (cover) — clean style:

python3 skills/instagram-photo-text-overlay/scripts/overlay.py \
  --input /tmp/ig-carousel/{dest-slug}-enhanced.jpg \
  --output /tmp/ig-carousel/slide-1.jpg \
  --title "{N} Day {DESTINATION} Itinerary Highlights" \
  --style clean --watermark "tabiji.ai"

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

python3 skills/instagram-photo-text-overlay/scripts/overlay.py \
  --input /tmp/ig-carousel/{slug}-enhanced.jpg \
  --output /tmp/ig-carousel/slide-{N}.jpg \
  --title "{ATTRACTION}" \
  --quote "{Specific insider tip about THIS attraction — must directly reference the place in the title, not a generic travel tip}" \
  --author "tabiji.ai" \
  --style quote --watermark "tabiji.ai"

Output: 6 overlay images at /tmp/ig-carousel/slide-{1-6}.jpg

Sub-agent 3: Publish to Instagram

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. Create carousel item containers (one per slide):

curl -s -X POST "https://graph.facebook.com/v21.0/${IG_USER}/media" \
  -d "image_url=${PUBLIC_URL}" \
  -d "is_carousel_item=true" \
  -d "access_token=${IG_TOKEN}"

1. Create carousel container with all children + caption:

curl -s -X POST "https://graph.facebook.com/v21.0/${IG_USER}/media" \
  --data-urlencode "caption=${CAPTION}" \
  -d "media_type=CAROUSEL" \
  -d "children=${CHILD_IDS}" \
  -d "access_token=${IG_TOKEN}"

1. Publish:

curl -s -X POST "https://graph.facebook.com/v21.0/${IG_USER}/media_publish" \
  -d "creation_id=${CAROUSEL_ID}" \
  -d "access_token=${IG_TOKEN}"

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

If media_publish returns a POST_ID, get the permalink directly:

curl -s "https://graph.facebook.com/v21.0/${POST_ID}?fields=permalink&access_token=${IG_TOKEN}"

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}"

If the most recent post timestamp is within the last few minutes, the publish likely succeeded — grab that permalink.

1. 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

Also clean up local temp files:

rm -rf /tmp/ig-carousel/

Output: Instagram post URL

Instagram API Auth

Keys from macOS Keychain:

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

Topaz API Auth

Key from macOS Keychain:

• topaz-api-key — Topaz Labs API key

Caption Template

🇲🇾 {N} Nights in {Destination} — {Itinerary Subtitle}

{One-line hook about the trip}

📍 Swipe through our top 5 picks:
1. {Attraction 1} — {one-line reason}
2. {Attraction 2} — {one-line reason}
3. {Attraction 3} — {one-line reason}
4. {Attraction 4} — {one-line reason}
5. {Attraction 5} — {one-line reason}

Full free itinerary with tips, prices & Reddit recs 👉 {ITINERARY_URL}

💬 {PROVOCATIVE_QUESTION — e.g. "Is 5 nights enough for {Destination} or do you need more?" or "What's the one thing most tourists get wrong about {Destination}?"}

#{destination_hashtag} #{country} #travelitinerary #foodietravel #southeastasia #asiatravel #travelguide #tabiji

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.