Competitor video ad remade for your brand

Competitor video ad remade for your brand

An n8n workflow. You send a competitor's video ad, a photo of your product and your brand's website through a form.
Gemini breaks the ad down shot by shot, Firecrawl reads your site, and Gemini writes brand notes and a prompt for a
12-second vertical ad with your product in it. The last step sends that prompt to OpenAI's Sora 2.

Read this first: OpenAI removed Sora 2 and its Videos API on 24 September 2026, so the five nodes that make the video
no longer work as imported. Everything before them still runs. See "The video step" below.

What is inside

• workflow.json: the workflow as n8n exports it, 25 nodes, one of them a sticky note that holds only the workflow's
  title. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The form's website field showed a real brand's address as its example. It now shows https://www.example.com/.
• Nothing else. The file has no Code node, and its expressions needed no repair.

Before you start

• An n8n instance, on n8n Cloud or your own server. The Edit Image node needs GraphicsMagick on a server of your own,
  which n8n's Docker image already has.
• A Google AI Studio key for the Gemini API, with access to gemini-2.5-pro.
• A Firecrawl account and its API key.
• An OpenAI API key, only if you still want the original video nodes, or another video service (see below).
• The rival ad as an MP4 file, your product photo and your site's address.

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 with your key, and select it in gemini-2.5-pro.
3. Add a Header Auth credential named x-goog-api-key, with the same key as its value. Select it in analyze_video_ad,
   which sends the ad straight to Google's API.
4. Add a Bearer Auth credential with your Firecrawl key. Select it in fetch_sora2_prompting_guide and
   fetch_site_details.
5. Decide on the video step (below). To keep the original nodes, add an OpenAI credential and select it in
   generate_sora2_video, get_video_status and get_video.
6. Save, open the form's test address and send an ad, a product photo and a web address.
7. Press Publish, then share the form's production address.

The form confirms at once, and the run goes on behind it. Open the run under Executions to read the shot breakdown, the
brand notes and the prompt, and to get the video when the last step works.

How a run goes

1. generate_video_trigger1 is the form, titled Sora 2 Video Ad Form, with three fields: Ad Video File,
   Product Image and Brand Website URL.
2. convert_video_to_base_64 turns the ad into text. set_ad_analysis_prompt holds the instructions for a shot list, and
   analyze_video_ad sends both to gemini-2.5-pro in one request, marked as video/mp4.
3. set_ad_analysis_result keeps Gemini's answer: an overall style, the core message, then each shot with its timing,
   visuals, camera, lighting, dialogue, sound and on-screen text.
4. fetch_sora2_prompting_guide has Firecrawl read OpenAI's Sora 2 prompting guide from its cookbook, and
   fetch_site_details has it read your website. Both carry on if Firecrawl fails.
5. build_guidelines_prompt and write_guidelines turn the site into brand notes in seven sections, from the brand's
   essence to its advertising angles.
6. build_sora2_prompt and write_sora2_prompt combine the shot list, the brand notes and the guide into one prompt for
   a 12-second ad with no on-screen text, and set_sora2_prompt_result keeps it.
7. Beside that, convert_image_to_base_64 reads your product photo. merge and aggregate put it and the prompt in one
   item, and convert_base64_image_to_binary and Edit Image turn the photo into a 720 by 1280 JPEG.
8. generate_sora2_video posts the prompt and the photo to OpenAI's /v1/videos with sora-2, 12 seconds, at the same size.
   delay waits 15 seconds, get_video_status asks again, and check_status loops until the status is completed.
   get_video then downloads the file.

gemini-2.5-pro is the model of both writing chains, and analyze_video_ad calls the same model by its address.

The video step

The prompt the workflow writes is in set_sora2_prompt_result, and the resized photo leaves Edit Image as a file
named data. To get a video again, replace generate_sora2_video, delay, get_video_status, check_status and
get_video with nodes for a video model that is still offered and takes a prompt and a starting picture. Veo is one
option, through Google's Gemini API or a reseller such as Kie.ai. Check its accepted lengths and sizes: the prompt is
written for 12 seconds, and the photo is sized for a vertical frame.

Change the look

• Length: the 12-second rule in build_sora2_prompt, and seconds in generate_sora2_video. Change both together.
• Frame: size in generate_sora2_video and the width and height in Edit Image.
• The brand notes: the seven sections in build_guidelines_prompt.
• The shot list: the format in set_ad_analysis_prompt.
• The ad's rules: the process and output rules in build_sora2_prompt, which ask for exact product looks, no other
  products and no text on screen.
• The form: the title and fields in generate_video_trigger1. If you rename a file field, change the node that reads it:
  Ad_Video_File in convert_video_to_base_64 and Product_Image in convert_image_to_base_64.

Good to know

• Sora 2: OpenAI's deprecations page, read on 2 November 2026, lists sora-2 and the Videos API as removed on
  24 September 2026, with no replacement. As imported, generate_sora2_video fails, so the run stops there.
• The brand notes see only your site's title. build_guidelines_prompt inserts $json.data.metadata.title, not the page
  text Firecrawl returns in $json.data.json.content, so Gemini guesses the rest. Add the content expression under the
  title line to use the whole page.
• The prompt writer never sees your photo. build_sora2_prompt asks it to study the product image, but only text reaches
  that chain. The photo goes to the video model alone, as the starting picture.
• The prompt asks for detailed descriptions of hands holding the product, then ends by asking to avoid showing hands.
  Pick one and edit the other out if the results come back mixed.
• Edit Image resizes the photo without keeping its proportions. Crop it to 9:16 first, or the product
  comes out stretched.
• analyze_video_ad sends the ad inside a single request and labels it video/mp4. Use a short MP4, since Google limits
  the size of a request that carries a file inline.
• Product Image leaves Multiple Files at n8n's default, which is on. Send one photo: a second one gets a numbered name
  that convert_image_to_base_64 does not read.
• check_status waits for completed only, so a failed video would be checked every 15 seconds without end.
• Each run costs Gemini calls for the analysis and two writing chains, and two Firecrawl scrapes, one of them the same
  cookbook page every time.
• Once the workflow is published, anyone with the form's address can start a run on your keys.
• The competitor's ad goes to Google, your website's address to Firecrawl. A remade ad can still copy the original too
  closely: change the idea, the words and the look enough to be your own, and check the rules on comparative and copied
  advertising where you sell.