wan-3-0-product-video

Wan 3.0 Product Video

Wan 3.0, Alibaba's video model, renders sound together with the picture: the hiss of a can, a candle wick, steam off a cup arrive in the same file, at the same price as a silent clip. This skill is built around that. You (Claude) pick a product whose sound helps sell it, write one brief that covers the camera move and what the viewer hears, and the Wireflow app skill-wan-3-0-product-video renders it from the seller's own photo.

App contract

If product_photo is left out, the app falls back to its saved demo candle. Always send it.

A. Connection check

Search your tools for names ending in run_app, get_execution, get_credit_balance, whoami, create_upload_link and get_upload_link.

Not connected? Say in one sentence that rendering needs the Wireflow connector, then:

• With this plugin: /mcp → wireflow → Authenticate.
• Plain Claude Code: claude mcp add --transport http wireflow https://www.wireflow.ai/api/mcp, then /mcp.
• Claude.ai or Desktop: custom connector, URL https://www.wireflow.ai/api/mcp.

Then hand over the free part: two or three shot and sound briefs written for their product, using reference/shot-and-sound.md. They work in any start-frame video tool that makes audio. Stop there.

B. Is sound the point?

Ask what the product is and where the clip will play. Then be straight with them:

• Wan 3.0 earns its price when the product or its setting makes a sound people recognise: fizz, pour, steam, a flame, a click, a zip, rain, a busy cafe. Product pages with autoplay muted still benefit on social, where sound is on.
• It is the wrong buy for silent products in silent settings (jewelry on velvet, a print on a wall). Point them to the ecommerce-product-video skill: a start-frame clip from a cheaper model, where the sound is an afterthought you would mute anyway.

C. The start frame

The photo is the first frame and sets the shape:

• Square photo in, square clip out (our test). Use a vertical photo for Reels and TikTok, a wide one for site banners.
• Sharp, real product, label readable. A clean setting with one or two props that can make the sound (a glass, a cup, a shelf by a window) gives Wan something to work with.
• Local file: create_upload_link, send the link, then get_upload_link when they have uploaded, and use that URL.

If their only photo is a cluttered phone shot, suggest making a clean scene first with the product-photography skill.

D. Write the shot and sound brief

One or two sentences, in this order: the move, what moves, the sound of that motion, the room. Examples and a table by product type: reference/shot-and-sound.md.

Slow push-in toward the lit candle on the stone shelf, the flame flickers gently and the wick crackles softly, warm afternoon light drifts across the wall, quiet room tone.

Rules that keep the clip clean:

• One move, slow. Wan follows the start frame closely; fast moves warp labels.
• The sound must come from something in the shot or a believable room. A crackle with no flame in view sounds wrong.
• No music, no speech, no on-screen text in the brief. Music is better added later in an editor.
• Only actions the product really does, and only once.

Write two options and let them pick.

E. Quote and confirm

Call whoami and get_credit_balance, then: "1 Wan 3.0 clip, 5 seconds with sound: 58 credits. Balance: 2,100." For a set, multiply. A short balance or a free plan (Wan 3.0 needs a paid plan): stop and say so. Nothing runs until they say yes to that plan and that cost.

F. Render

{ "slug": "skill-wan-3-0-product-video", "inputs": { "product_photo": "https://...", "shot_and_sound": "Slow push-in toward ..." } }

Poll get_execution every 5 seconds for the first minute, then every 15. Tell them it is rendering: our test took about 2.5 minutes, and Wan's prompt expansion is part of that, so allow longer when Wireflow is busy. COMPLETED gives the MP4 URL. FAILED, CANCELLED or TIMEOUT: stop and quote the error.

Save to ./wireflow-outputs/<YYYY-MM-DD>/wan-<product>-<move>.mp4 with curl -sSL -o or PowerShell Invoke-WebRequest -OutFile.

G. Check picture and sound

With ffmpeg:

ffmpeg -v error -i wan-candle-push-in.mp4 -vf "fps=1,scale=240:-2,tile=5x1" -frames:v 1 frames.jpg
ffmpeg -v info -i wan-candle-push-in.mp4 -af volumedetect -f null -

• Frames: label readable in every frame, shape and color stable, no second product, no warping at the edges.
• Sound: mean_volume well above -60 dB means there is real audio. Frames cannot tell you whether the sound matches the motion, so ask them to listen once with headphones.

A redo is another full run (58 credits). Ask first and change one thing: the move, the sound, or the photo.

H. Hand over

Give the file path, the CDN link and the credits used (balance before minus after). Then offer:

• The same move and sound on the rest of the range, quoted as one batch.
• The board in Wireflow, to change length (Wan 3.0 goes from 2 to 30 seconds), resolution or add an end frame: https://www.wireflow.ai/flow/cmva7a6qz000jl90697edudo9?ref=skill-wan-3-0-product-video&utm_source=github&utm_medium=skill&utm_campaign=wan-3-0-product-video

Guardrails

• No run before a yes to the exact plan and cost, and no silent re-runs.
• Never show the product doing something it cannot do, and add no claims, prices or ratings.
• Use only photos the person owns or has the right to use.
• Keep keys and passwords out of inputs and files.
• When Wireflow refuses or a tool errors, quote the exact message and offer send_feedback if it exists.