Property Marketing Kit

Property Marketing Kit

Turn a single listing URL (or a search) into a complete, owner-ready marketing kit:

listing → scrape photos & details → organized kit folder → enhanced hero photos
        → cinematic tour video + vertical UGC reel → property website → email/handoff

This skill is the playbook behind the Higgsfield "analyze a listing, build the assets the owner never had, then pitch them" workflow. A complete worked example ships in assets/sample-kit/ — real enhanced photos and a real tour reel produced by this exact pipeline for a Clearwater Beach, FL beachfront estate. Use it as the quality bar.

Prerequisites (check first, every run)

1. Higgsfield MCP — required for image/video generation. Tools are named

mcp__*__generate_image, mcp__*__generate_video, mcp__*__models_explore, mcp__*__media_import_url, mcp__*__job_display. If you can't find them, run ToolSearch with generate_video higgsfield. If still absent, tell the user the Higgsfield MCP must be connected and stop.

1. Firecrawl — preferred scraper. See references/firecrawl-setup.md to verify it's

installed and, if not, the one-liner to add it. A headless-browser fallback (Playwright/Chrome MCP) is built in and works without Firecrawl — see references/scraping-listings.md. Never block the run just because Firecrawl is missing; confirm with the user, then fall back.

> Preflight every Higgsfield generation with get_cost: true and tell the user the credit > total before spending. A basic run ≈ 25–30 credits; a full kit with a stitched multi-shot > tour + host walkthrough ≈ 60–90 credits. Concurrency: some plans cap concurrent jobs > (pro = 3). Submit generations in waves of ≤3 and poll job_display before sending more — > the API rejects overflow with a rate-limit error.

The pipeline

Work through these in order. Each step has a dedicated reference — read it before doing the step.

1 — Intake & scrape → references/scraping-listings.md

• Get the listing URL (or run a search per the user's criteria and pick one).
• Scrape with Firecrawl; if it can't render the gallery (most modern listing sites are JS

apps that lazy-load photos), use the Playwright DOM-harvest technique in the reference.

• Critical: listing galleries often lazy-load other listings' photos into the DOM.

Verify every image actually belongs to the target property (open a few; drop strays).

• Extract: name, location, beds/baths, sleeps/sqft, view, highlights, full description,

and the direct full-resolution image URLs.

2 — Organize the kit folder → run scripts/new_kit.sh

Create a self-contained folder (default under ./property-marketing-kits//):

01-source-photos/     downloaded originals (photo-01.jpg …)
02-enhanced-photos/   Higgsfield-enhanced heroes
03-tour-video/        tour + vertical reels
04-website/           index.html + assets/
05-marketing-copy/    copy.md (headlines, captions, owner email)
listing.json          structured listing data + a labeled photo manifest

Download originals with scripts/download_images.sh. Then look at every photo (Read them) and write listing.json with a one-line label per image so later steps pick the right shots.

3 — Enhance hero photos → references/higgsfield-models.md

• Pick 2–3 strongest shots (a twilight/exterior hero, a pool/outdoor hero, a key interior).
• media_import_url each chosen original → get media_id.
• generate_image with nano_banana_pro (resolution 2k, the listing's best aspect

ratio, usually 16:9), passing the original as medias[].role:"image". Prompt = the real-estate enhancement recipe in the reference (relight, sky, crisp water, declutter, keep architecture identical). Poll with job_display; download results to 02-.

4 — Tour videos → references/higgsfield-models.md + references/video-stitching.md

Generate clips, then stitch into full tours. Submit in waves of ≤3 (concurrency cap).

• Hero tour clip (16:9): veo3_1 (8s, with audio), start_image = the enhanced hero's

job id. Slow cinematic push-in, no warping, ambient sound.

• Complex multi-shot tour: animate 4–6 enhanced scenes (aerial, exterior, pool, great room,

firepit, sunset) with cinematic_studio_video_v2 (~5cr each), then stitch with ffmpeg into one property-tour-cinematic.mp4 (~30–40s). Normalize → concat per the stitching ref.

• Vertical UGC reel (9:16): cinematic_studio_video_v2 (5s, genre:"intimate"), a

lifestyle shot.

• If Higgsfield returns a preset_recommendation, use it or re-send with declined_preset_id.

Download all clips to 03-tour-video/clips/, finished tours to 03-tour-video/.

4b — Host / agent walkthrough (optional) → references/host-walkthrough.md

Put a real presenter "showing people around" the home — often the user themselves.

• Get their photo as a media_id (media_upload_widget, or reuse one from show_medias).
• Composite them into 3–4 real rooms with nano_banana_pro (two inputs: room + person;

"preserve exact facial identity"). Verify each face (Read it); regen drifts.

• Animate each composite (veo3_1 for the hero shot, cinematic_studio_video_v2 for the

rest) and stitch into host-tour-.mp4. Keep a 9:16 cut for socials.

5 — Build the website → references/website-template.md

• Copy assets/website-template.html into 04-website/index.html and drop chosen images

into 04-website/assets/img/ and the two clips into assets/video/.

• Fill the placeholders ({{PROPERTY_NAME}}, beds/baths/sleeps, highlights, copy). The hero

is a muted autoplay loop of the tour video with the enhanced photo as poster.

• Preview it (the Launch panel, or open index.html). Keep the AI-asset disclosure in the footer.

6 — Marketing copy → write 05-marketing-copy/copy.md

Headlines, long description, social captions (IG/FB/TikTok), and the owner-outreach email. Template lives in references/website-template.md and the sample kit's copy.md.

7 — Handoff / email / autopilot → references/autopilot-routine.md

• Default handoff = the organized folder (the user reviews it). Surface the website file.
• If the user wants it emailed or running on a schedule, use the routine in the reference

(Gmail draft + the /schedule skill). Never auto-send outreach without approval.

Guardrails

• Honesty: these are AI-enhanced/AI-generated assets. Keep the footer disclosure on the

site and never imply photos are unretouched originals.

• Respect sources: scrape only public listing pages; don't defeat paywalls/logins.
• Cost: always preflight with get_cost and report the total before generating.
• Verify provenance of every scraped photo (see step 1) before it lands in the kit.

See references/ for the full detail on each step and assets/sample-kit/ for the finished bar.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Samin12
• Source: Samin12/property-marketing-kit
• License: MIT

Install and usage instructions live in the source repository linked above.