🎬 Generate competitor-style image ads from your product photos with Gemini

⚡ 273 views · 🎬 Content Creation & Video

💡 Pro Tip — For competitive intelligence, ScraperNode can automate the data collection — Yelp reviews, Glassdoor company data, and Crunchbase profiles all return structured JSON you can feed straight into this workflow.

Description

Image Ad Cloner AI Agent

This n8n workflow automates “cloning” a competitor ad style and re-generating it using your product image. A user uploads your product image and provides a competitor ad URL via a form trigger; the workflow downloads both images, converts them to inline/base64 parts, builds a detailed generation prompt, sends the request to a generative-vision model (Gemini-style endpoints), waits for the result, converts the output back to a file and returns it. Use-cases: rapid ad iteration, creative A/B variants, and scaled ad re-skins.

How it works (step-by-step)

form_trigger — user submits: competitor image URL, their product file, and optional changes.
convert_product_image_to_base64 — converts uploaded product file to base64 inline data.
download_image — fetches the competitor ad image from the provided URL.
convert_ad_image_to_base64 — converts downloaded competitor image to base64 inline data.
build_prompt — assembles a careful prompt (handles partial text, labels, placements, CTA swaps, additional changes).
generate_ad_image_prompt → generate_ad_image — two HTTP Request nodes call generative endpoints (first to prepare prompt/content and then flash-image generation). These include inline image data and model-specific parameters.
Wait — allows async generation to complete (workflow uses a wait/webhook pattern).
set_result → get_image — extract image result, convert binary to file for downstream use (download or return to the requester).
Sticky Note — visual diagram / docs inside workflow.

Quick Setup Guide

👉 Demo & Setup Video 👉 Course

Nodes of interest

form_trigger (n8n-nodes-base.formTrigger) — entry point; validates inputs.
convert_product_image_to_base64 & convert_ad_image_to_base64 (extractFromFile) — convert uploaded/downloaded binaries to inline/base64 data for the model.
download_image (httpRequest) — fetches competitor image URL. Add robust URL validation.
build_prompt (set) — builds the natural-language prompt that instructs the model how to replace branding, packaging, and CTAs.
generate_ad_image_prompt & generate_ad_image (httpRequest) — POST to generative APIs (examples use Gemini endpoints). Have retry/backoff settings.
Wait — used to allow model generation to complete before extracting results.
set_result / get_image / convertToFile — assemble the final binary file to deliver to the user.

What you’ll need (credentials & permissions)

n8n account with Workflow active toggle & webhook exposure (or self-hosted endpoint).
HTTP credential(s) for your generative model provider:
- Header auth credential for any required custom headers (httpHeaderAuth).
- Bearer token credential for the model API (httpBearerAuth).
- (Optional) Generic HTTP auth if using proxies or intermediate APIs.
Storage (optional): S3 or other if you want to persist generated images.
(Recommended) An OCR or moderation service API key if you plan to detect/remove copyrighted logos or PII.
Ensure your hosting domain allows outbound HTTPS to the model endpoints and that webhook URLs are reachable.

Recommended settings & best practices

Input validation: enforce allowed mime types (jpg, png, webp), max file size (e.g., 5–10 MB), and sanitize competitor URL domain whitelist.
Retries & backoff: use retryOnFail and waitBetweenTries for httpRequest nodes (example uses retryOnFail: true and waitBetweenTries: 5000). Limit retries to avoid abuse.
Rate limits & quotas: guard calls to the model API with rate-limiting / queueing to avoid hitting provider quotas.
Ethics & copyright check: run OCR/logo-detect and flag requests that attempt to reproduce trademarked content; require human approval before creating ads that replicate competitor branding.
Partial-text handling: include logic in the prompt (and optionally OCR pre-check) to identify fragment text on packaging and replace it correctly with your brand text.
Binary handling: always set and preserve correct mime_type in inline data (png/jpeg) to avoid corrupt outputs.
Webhooks & timeouts: set a sensible Wait time and webhook timeout; provide progress/error messages back to the requester.
Logging & monitoring: capture model responses and node errors to troubleshoot prompt issues and image artifacts. Obfuscate tokens in logs.
Security: store API keys in n8n credentials (never as plain text in Set nodes). Use least privilege.

Customization ideas

Add an OCR node (e.g., Tesseract or cloud OCR) prior to build_prompt to detect and list text to replace automatically.
Add a logo mask step to blur or remove competitor logos before feeding to the generator.
Generate multiple variants per request (different CTAs, color palettes) and save them for A/B testing.
Add human-in-the-loop: store generated images in a review queue and send approval emails/webhooks before final download.
Integrate analytics: tag generated ads and push metadata (prompt, model used, timestamp) to a database for later analysis.
Replace Gemini endpoints with other providers or allow a model selector UI option.
Add automatic compliance moderation filters and a “safe mode” toggle that refuses disallowed transformations.