Fast-cut 15-shot product commercial from one photo

Fast-cut 15-shot product commercial from one photo

An n8n workflow that turns a product photo and the product's name into a 15-second commercial, cut to one shot a second.
Gemini describes the product, then writes a storyboard of fifteen shots and a short voiceover that ends on the name and
a tagline. Cloudinary hosts the photo, and Seedance 2.0 on Kie.ai renders the clip from the storyboard with the photo as
its reference.

What is inside

• workflow.json: the workflow as n8n exports it, ten nodes and no sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The video model. The original copy asked Kie.ai for sora-2-image-to-video. OpenAI's deprecations page lists Sora 2
  as removed from its API on 24 September 2026. On 2 November 2026, Kie.ai's documentation page for that model
  answered 404 Not Found, and its index listed no Sora model. The call could no longer run, so submitTask now asks
  for bytedance/seedance-2 on the same jobs/createTask endpoint. The original's choices are carried over:
• n_frames: "15" became duration: 15. Seedance takes 4 to 15 seconds.
• The two frames became 9:16 and 16:9. The original also mapped two HD choices the form never offers, so they
  are gone.
• image_urls became reference_image_urls, so the photo guides every shot instead of being the first frame.
• remove_watermark is gone, since Seedance has no such field. resolution is set to 720p and generate_audio
  to true, the model's own defaults written out.
• The Kie.ai key in the headers of submitTask and checkStatus was Bearer <token>. It now reads
  Bearer <YOUR_KIE_API_KEY>.
• The Gemini credential is gone from both Gemini nodes.

checkStatus, Switch and downloadAdVideo are unchanged. Kie.ai's task page lists the same states and the same
resultJson with resultUrls for every model on that endpoint.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Gemini API key from Google AI Studio.
• A Cloudinary account, its cloud name and an upload preset whose signing mode is Unsigned.
• A Kie.ai API key, with credit for one Seedance clip on each run.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Google Gemini (PaLM) API credential and select it in analyzeProductImage and Message a model.
3. If your Gemini key is new, pick models/gemini-3.8-flash as the model in both of those nodes. Google's deprecations
   page, read on 2 November 2026, sets no shutdown date for gemini-2.5-flash, but it limits the 2.5 models to
   accounts that have used them before and names Gemini 3.8 Flash for new projects.
4. In uploadImageCloudinary, put your cloud name in the address where it says <YOUR_CLOUD_NAME>. Swap
   YOUR_PRESET_NAME for your preset and YOUR_FOLDER_NAME for a folder.
5. In submitTask and checkStatus, replace <YOUR_KIE_API_KEY> in the Authorization header with your key. The
   brackets go too, so the header reads Bearer, a space and the key.
6. Save, open the test address of On form submission and send one photo. Press Publish when the clip looks right.

How a run goes

1. On form submission shows the Commercial Ad Agent form: Product name, Upload product image (JPG, PNG or WebP)
   and Screen resolution (Portrait or a wide frame), all three required. It confirms at once and the run goes on behind it.
2. analyzeProductImage asks gemini-2.5-flash for a one-line description of the product in the photo.
3. Message a model sends the product name and that line to gemini-2.5-flash with a long system message. It plays a
   director of photography and writes fifteen shots of one second each: a dark tease, close-ups of the material, a
   person using the product, then the name and a tagline. Each shot has a visual prompt and an audio line.
4. getBinaryFile takes the photo back from the form, since the Gemini nodes passed on text only.
5. uploadImageCloudinary uploads it with your unsigned preset and gets its secure_url.
6. submitTask sends the whole storyboard, the photo's address, the frame and the length to Kie.ai.
7. Wait pauses five minutes. It is version 1.1 with the unit set to minutes and no amount, so it takes the version's
   default amount of 5. checkStatus then asks for the task.
8. Switch reads data.state. success goes on to downloadAdVideo, and waiting, queuing or generating go back
   to Wait. fail has no node after it.
9. downloadAdVideo fetches the clip from the first address in resultUrls. It stays in the run as a file.

Change the look

• The arc, the pace and the voiceover: the system message of Message a model. The ending rule asks for the name and a
  tagline. The form has no tagline field, so the model writes one. To use your own, add a field to the form and a
  Tagline: line to the text of Message a model.
• The look of every shot: the tech specs line in the same message, which is added to every visual prompt.
• Frame, length and quality: aspect_ratio, duration and resolution in the body of submitTask. Seedance takes
  480p, 720p, 1080p or 4k.
• Sound: generate_audio in the same body.
• How often it checks: Wait. Set an amount of 1 or 2 for a shorter wait between checks.

Good to know

• Kie.ai bills each clip by its length, at the prices on its pricing page. The two Gemini calls cost
  little next to it. Try one photo before a batch.
• Once the workflow is published, anyone with the form's address can start a render on your Kie.ai credit.
• An unsigned Cloudinary preset leaves the photo readable by anyone with its address.
• A render that fails ends the run without a message. Add a node on the fail output of Switch to be told.
• Seedance makes sound with the picture by default. How closely it voices the storyboard's audio lines is up to the
  model, so listen before you post.
• The photo and the product name go to Google, Cloudinary and Kie.ai.
• The Kie.ai key sits in two headers, so take it out before you share the workflow.