CGI product ad video built from a hero frame

What it is

This n8n workflow turns a single product photo into a 10-second video ad that looks rendered in 3D. Thirty nodes run in one line. Three sticky notes name the stages. A form takes the photo, and Dropbox keeps it behind a public link. OpenAI describes the product, picks an ad concept and writes a detailed prompt for one still, the hero frame. A Flux model on Runware paints that frame with your photo as its reference. Kling, also on Runware, then animates it. The run ends with the video’s address.

1. A form takes one product photo
2. Dropbox keeps it behind a public link
3. OpenAI describes it and plans a scene
4. Flux on Runware paints the hero frame
5. Kling on Runware animates it for 10 seconds

Who it suits

Shops and brands that want scenes no one could shoot cheaply, such as a perfume bottle standing on a frozen lake. You send only the photo, and the workflow chooses the idea itself. That makes it a source of quick drafts more than a tool for a planned campaign. Skip it when every frame must show the product exactly as photographed. The hero frame is a new picture, made from yours. Your photo itself never appears.

Where to use it

Any n8n that can reach OpenAI, Dropbox and Runware will run it. The finished video stays at an address on Runware, so plan where it goes next.

• n8n Cloud: import the file into a new workflow there
• Your own n8n server: the same file, the same steps
• Ads and social posts: download the clip from its address and post it as an ad, a Reel or a Short
• A drive or a sheet: add a node after Edit Fields1 to keep the address or save the file

Setup

Get it saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. Its author puts setup at about 12 minutes.

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Add an OpenAI credential and select it in Analyze image, OpenAI Chat Model, OpenAI Chat Model1 and OpenAI Chat Model2.
3. In Dropbox’s App Console, give your app the files.content.write, sharing.read and sharing.write permissions, then generate an access token.
4. Add a Dropbox credential with that token and select it in Upload a file.
5. In HTTP Request1 and HTTP Request6, replace <YOUR_DROPBOX_ACCESS_TOKEN> after Bearer with the same token. The brackets go too.
6. In Flux Image, kling image to video post and kling imag to video get status, replace <YOUR_RUNWARE_API_KEY> the same way with your Runware key.
7. Save, send a test photo through the form’s test address, then press Publish.

Tokens from the App Console are short-lived now, so a pasted one stops working after a while. For a setup that lasts, give Upload a file a Dropbox OAuth2 credential. Then set both HTTP Request nodes’ Authentication to it and delete their Authorization header.

How to use it

Open the form from On form submission, attach one product photo in upload product image and send it. The form confirms at once. Behind it the run needs at least three minutes and forty seconds, because it waits ten seconds after the frame and three and a half minutes after asking for the video. Open the run under Executions and read finalVideoURL in Edit Fields1.

Example

A photo of a matte black water bottle on a white table goes through the form. The description from Analyze image runs from the label and the lid to the colors and the size. One concept comes back from Ad Ideas: an emoji and a title in curly quotes, then a short scene on a second line. From both, Image Prompt writes the hero-frame prompt, and its parser hands it on under one key:

{
  "Ad Concept 1": "Ultra-realistic 8K CGI render of … --ar 16:9"
}

Flux draws the frame from that prompt and your Dropbox photo. Kling then gets a body of this shape, trimmed here:

[
  {
    "taskType": "videoInference",
    "model": "klingai:5@1",
    "positivePrompt": "<the one-line motion prompt>",
    "frameImages": [{ "inputImage": "<the hero frame’s address>", "frame": "first" }],
    "duration": 10,
    "width": 1280,
    "height": 720,
    "deliveryMethod": "async"
  }
]

Three and a half minutes later, one call to Runware fetches the video’s address. Nothing is downloaded.

For newcomers and experts

If you are new. n8n draws a workflow as boxes joined by lines, and each box, a node, does one job. A credential is a key n8n keeps for you and hands to the node you pick it in. Send a photo from the form’s test address first. Each node then shows what it got and what it sent, so a wrong token or a missing Dropbox permission turns up where it failed.

For experts. Three OpenAI models do the writing. Runware makes the picture and the video, both through one endpoint.

The frame is square and the video is not, so Kling starts from a square picture. Dropbox’s link ends in raw=1 after Code4, which lets Runware read the file itself.

Good to know

Each run costs money on two services. Two steps can fail quietly.

• Billing: OpenAI bills four model calls a run, and Runware bills the frame and the video
• The concept: Ad Ideas has no output parser, so its answer arrives as text, and Image Prompt looks up Ad Concept 1 inside it as if it were JSON. The scene then rests on the description alone. Switch on Require Specific Output Format in Ad Ideas and attach a parser with the same example as Structured Output Parser1
• One status check: a video still rendering after the wait leaves finalVideoURL empty, so lengthen Wait or loop back to it until a videoURL arrives
• Quotes: both prompts sit between quotes in the Runware bodies, so a straight double quote breaks the request. Wrap the expression in JSON.stringify in place of the quotes
• The original copy: its Upload a file cleaned names with /s+/g, which hits the letter s, and the line breaks in the Ad Ideas prompt had lost their backslashes. This copy has them back
• Your photo: it stays at the top of your Dropbox with a public link, and it also goes to OpenAI and Runware
• Open form: once the workflow is published, anyone with the form’s address can start a run on your keys, so turn on the form trigger’s basic authentication or keep the address to yourself
• Credentials: the JSON holds none, and nothing from its author’s account