BEFORE YOU START
What you need
This runs inside Claude Code, Anthropic's terminal-based coding tool, not the regular claude.ai chat window. The reason: the last step stitches the video clips together with ffmpeg on your own machine, and only Claude Code can run that.
- Claude Code installed (claude.ai/code) — free to install, pay-as-you-go usage
- ffmpeg installed on your computer (ffmpeg.org — required for the stitching step)
- A Higgsfield account (higgsfield.ai) — this is what actually generates the video clips
- An Apify account (apify.com) — this is what pulls the listing photos automatically from a Zillow link
Important: Both Higgsfield and Apify have free tiers to get your first walkthrough done. Full videos after that run on credits/usage, a few dollars per property.
STEP 1 / CONNECT HIGGSFIELD
This is what generates the video clips
Higgsfield ships an official CLI that wires it straight into Claude Code.
npm i -g @higgsfield/cli
higgsfield auth login
npx skills add higgsfield-ai/skillsImportant: The middle command opens your browser to sign into your Higgsfield account. No API key to copy-paste.
STEP 2 / CONNECT APIFY
This is what pulls the listing photos and details automatically
Inside Claude Code, install the official Apify plugin.
1
Add the marketplace
Run /plugins in Claude Code, choose Marketplaces → Add Marketplace, and paste this repo:
https://github.com/apify/apify-claude-code-plugin2
Install and connect
Go to the Discover tab, install the "apify" plugin, then run /reload-plugins, then /mcp, select plugin:apify:apify, and choose Enable. It opens your browser to sign into Apify.
Important: No account yet? Sign up free at apify.com first, then run the steps above.
STEP 3 / COPY THIS PROMPT
This is the entire build. Paste it into Claude Code as one message.
Swap in your own Zillow listing URL where marked, and send it. This is the full recipe, not a summary of one.
I want a seamless AI walkthrough video from this Zillow listing: [PASTE YOUR ZILLOW LINK]
Do this exactly:
1. Scrape the listing with the Apify actor maxcopell/zillow-detail-scraper (startUrls = my link, propertyStatus = FOR_SALE). Pull the address, price, specs, and every photo URL.
2. Look at every photo. Drop floorplans, aerial/map shots, and near-duplicates. Pick 8 to 10 photos that each show a different room, in this walking order: exterior, entry, main living space, kitchen, stairs (if any), hallway, bedrooms, bathrooms, backyard. Don't just guess the order — for each pair of consecutive photos, check if you can actually see a real connection (a matching door, a sightline into the next room, a staircase that continues). If you can't find one, tell me and mark that transition as "not a real connection" instead of forcing it.
3. Import each of the 8-10 chosen photos into Higgsfield with media_import_url.
4. For every pair of consecutive photos, generate one clip with generate_video, model kling3_0, medias = [start_image: photo A, end_image: photo B], aspect_ratio 16:9, mode pro, duration 5, sound off. Write the motion prompt like this:
- Real connection: describe a forward walk through the specific door/staircase/opening you identified. Example: "Continuous forward walking motion through the open doorway, camera glides directly into the next room, smooth steadicam threshold reveal, natural light, no people."
- No real connection: describe a slow cinematic push with a dissolve-like quality instead, so it reads as intentional. Example: "Slow cinematic push forward with a graceful dissolve-like transition, camera glides forward into the next space, warm light, no people, elegant transition."
One camera move per clip. Slow, not fast.
5. Wait for every clip to finish rendering, then download them in order.
6. Before stitching: for every photo used twice in the chain (the end of one clip AND the start of the next), open that actual photo file again and confirm it really shows what you think it shows. If one is wrong, fix it and regenerate both clips that touch it.
7. Normalize every clip to the same resolution, fps, and codec, then stitch them together in order with ffmpeg into one final video.
8. Show me the final video file and the total Higgsfield credit cost.STEP 4 / RUN IT
Paste your link, then wait
Claude Code runs the whole thing on its own from there: scrape, curate, generate, check, fix anything wrong, stitch. A typical 8-10 room property takes roughly 10-20 minutes end to end, mostly video render time. You'll see it working through each step as it goes.
STEP 5 / WHAT YOU GET
One finished video file, ready to hand to an agent
A single MP4, camera gliding room to room with no hard cuts between them. It ships silent — add music or a voiceover in your editor afterward. Expect roughly $0.01 in Apify usage plus somewhere around 60-90 Higgsfield credits per property, depending on room count.
Call it a cinematic AI walkthrough when you sell it. It's AI-generated camera motion over real listing photos, not a 3D scan or a Matterport tour, and it shouldn't be sold as one.
Official Resources
KEEP LEARNING
One guide down. The full library is waiting.
Go back to the Dorian Drone resource library for the Part 107 roadmap, creator systems, drone workflows, and new field guides.
Browse All Free GuidesNo paywall. No fake results. No fluff.