Product photo to 360-degree turntable video

Product photo to 360-degree turntable video

An n8n workflow behind a one-field upload form. Send it a photo of a product, and it makes a 10-second video of that
same product turning a full circle on a white studio turntable. OpenAI describes the product and writes the prompts.
Runware renders a square hero frame with FLUX.1 Kontext [pro], then animates it with a Kling model. Dropbox holds the
photo and gives Runware a link to it.

It adds no scene, story or presenter. The product should look as it does in the photo, only turning.

Read "Models to check first" below before the first run: three model names in the file need a look.

What is inside

• workflow.json: the workflow as n8n exports it, 32 nodes, four of them sticky notes that label its four rows. It
  holds no credential.
• README.md: these notes.

What we changed from the original copy

• Its five Authorization headers held Bearer <API>. The two that call Dropbox, in HTTP Request and HTTP Request6,
  now say Bearer <YOUR_DROPBOX_TOKEN>. The three that call Runware, in Flux Image , kling image to video post1 and
  kling imag to video get status1, say Bearer <YOUR_RUNWARE_KEY>.
• The original copy's Upload a file path had lost a backslash. Its /s+/g turned every letter s in the file name into
  an underscore. It reads /\s+/g again, which turns spaces into underscores.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Dropbox account, and an app made in Dropbox's App Console (below).
• An OpenAI API key for a model that reads images.
• A Runware account with credit, and its API key.

Models to check first

Analyze image and OpenAI Chat Model1 name chatgpt-4o-latest. OpenAI's deprecations page, read on 2 November 2026,
lists it as shut down on 17 February 2026. Pick a current model in both, such as gpt-4.1-mini, which reads images too.

kling image to video post1 asks Runware for klingai:5@1. That ID was not in Runware's model list when we checked the
same day. Its Kling models then included klingai:6@0, KlingAI 2.5 Turbo Standard, which makes clips of 5 or 10
seconds at up to 720p. Runware's image-to-video guide also puts the start frame under inputs.frameImages now. It says
many models refuse width and height when a frame is given. If the call fails, move frameImages under inputs as
the guide shows, and take the two sizes out.

A Dropbox token

Upload a file is n8n's Dropbox node, set to take an access token. HTTP Request and HTTP Request6 call Dropbox's API
with a token in their headers. The quickest setup uses one token in all three:

1. In Dropbox's App Console, create an app with scoped access. App folder access is enough, and the photos then land in
   the app's own folder.
2. On its Permissions tab, tick files.content.write, sharing.read and sharing.write, and submit.
3. On its Settings tab, generate an access token.

Dropbox's OAuth guide says its access tokens are short lived, and that a token made in the App Console is for testing
with your own account. Expect to paste in a new one after a few hours.

For a setup that lasts, switch Upload a file to OAuth2 and give it n8n's Dropbox OAuth2 credential. That credential
asks Dropbox for files.content.write, files.content.read, sharing.read and account_info.read, which cannot make
the link in HTTP Request6. So give the two HTTP Request nodes a generic OAuth2 API credential for your app instead:

• Authorization URL https://www.dropbox.com/oauth2/authorize and Access Token URL
  https://api.dropboxapi.com/oauth2/token, with your app's key and secret
• Scope sharing.read sharing.write, and token_access_type=offline in the auth URI query parameters
• n8n's OAuth redirect URL added to the app's Redirect URIs in the App Console

Then delete their Authorization headers.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Make the Dropbox app and its token (above). Add a Dropbox credential with the token and select it in Upload a file.
3. In HTTP Request and HTTP Request6, replace <YOUR_DROPBOX_TOKEN> with the same token. The brackets go too, so the
   header reads Bearer, a space and the token.
4. Add an OpenAI credential and select it in Analyze image, OpenAI Chat Model1 and OpenAI Chat Model2. Change the
   model in the first two (above).
5. In Flux Image , kling image to video post1 and kling imag to video get status1, replace <YOUR_RUNWARE_KEY>
   with your Runware key, brackets included. The first of those names ends in a space.
6. Check the model in kling image to video post1 against Runware's list (above).
7. Save. Open On form submission, use its test address and upload a product photo. Wait about four minutes, then open
   the run and read finalVideoURL in Edit Fields3.
8. Press Publish when the video looks right. The form then has a production address you can share.

How a run goes

1. On form submission serves the Image trigger form, which has one file field. Code2 names each file it receives
   product_image, one item for each.
2. Upload a file saves the photo at the top of your Dropbox, or of the app's folder, with spaces in its name turned
   into underscores.
3. HTTP Request asks Dropbox for a link to that file. When If finds one, Code7 takes it. Otherwise
   HTTP Request6 makes a public link first. Code7 swaps dl=0 for raw=1, so the address serves the file itself.
4. HTTP Request3 downloads the photo from that link, and Code6 labels it a PNG named product.png.
5. Analyze image has OpenAI describe the product in detail: its label, material, logo, colors, shape and size.
6. Basic LLM Chain turns that description into one render prompt. Its rules ask for an exact copy of the product,
   centered on a glossy white turntable in a studio light box, with no stylizing. The prompt has to open with
   "Ultra-realistic 8K CGI render of", and its last words are --ar 1:1. Structured Output Parser1 holds the answer to one
   field, Ad Concept 1, and Edit Fields and Code pass it on as prompt.
7. AI Agent1, on gpt-4o-mini, writes the video prompt: one sentence of up to 40 words about a smooth 360-degree turn.
8. Code3 makes a task ID. Flux Image asks Runware for one 1024 by 1024 PNG from bfl:3@1, FLUX.1 Kontext [pro],
   with the Dropbox link as its reference image. Split Out1 takes the result, and Wait6 pauses 10 seconds.
9. Edit Fields2 keeps the frame's address under the name finalVideoURL, though it is still the picture. Code5
   makes a second task ID, and kling image to video post1 asks Runware for a 10-second, 1280 by 720 video with that
   frame first. Runware answers at once and renders in the background.
10. Wait1 pauses three and a half minutes. Then kling imag to video get status1 asks once for the result, and
    Edit Fields3 sets finalVideoURL to the video's address.

Change the look

• The turntable, backdrop and light: the rotation context at the end of the system message in Basic LLM Chain.
• The camera move: the prompt in AI Agent1.
• The video's length and size: duration, width and height in the body of kling image to video post1.
• The frame's size: width and height in Flux Image .
• The waits: Wait6 after the frame, and Wait1 before the one check on the video.

Good to know

• The workflow checks on the video once. If Runware has not finished after three and a half minutes, finalVideoURL
  comes out empty. Run kling imag to video get status1 again, or add an If that loops back to Wait1 until a
  videoURL appears.
• The video's address stays in the run. The person who sends the form sees only n8n's confirmation. Add a node that
  emails it, saves it or posts it if anyone else needs the video.
• Both prompts go into the JSON bodies as they are. A quote mark or a line break in a model's answer breaks the request.
  Writing {{ JSON.stringify($('Basic LLM Chain').item.json.output['Ad Concept 1']) }} in place of the quoted text in
  Flux Image fixes that, and the same works for the video prompt.
• The frame is square and the video request is 16:9, so the model may crop the frame or add space around it.
• --ar 1:1 comes from Midjourney's prompt syntax. FLUX reads it as plain text. The frame is square because of the
  width and height in Flux Image .
• Code6 labels the photo a PNG whatever it is. Upload a PNG or a JPEG.
• The Dropbox link is public, so anyone who has it can open the photo. Remove the link in Dropbox once the video is made.
• Anyone with the form's production address can start a run on your OpenAI and Runware credit. The form trigger can ask
  for a password with Basic Auth.
• Each photo costs three OpenAI calls, one Runware image and one Runware video. Runware prices each model on its own, so
  check before you choose one.
• The Dropbox token and the Runware key sit in headers, so take them out before you share the workflow.