Analyze Competitor YouTube Thumbnails with n8n + OpenAI Vision and Hand Off Approved Concepts to ThumbCanvas

Collect briefs and competitor URLs → extract public metadata and visual evidence → AI generates three original directions → human approval gate → ThumbCanvas link + Google Sheets audit

Who

YouTube creators, thumbnail designers, and content teams who want evidence-informed creative directions without copying competitors or pretending visual analysis guarantees CTR

You get

Enter competitor YouTube URLs and a planned video brief, get OpenAI Vision analysis of thumbnail evidence, generate three original directions, and after human approval produce a ThumbCanvas deep link and append an audit row to Google Sheets

You need

n8n, Google Sheets

Steps

1. Collect and Validate Brief

   Fill the n8n Form Trigger with planned video title, promise, audience, target keyword, brand palette, optional thumbnail text (max 4 words), and 2–3 competitor YouTube URLs. The Normalize and Validate Brief Code node normalizes fields, extracts strict 11-character video IDs, deduplicates, and validates. If validation fails, an HTTP 400 error page is returned with correction messages.

   Setup guidance · Source text。Overview Setup and Safety sticky note content field (overview and safety, verbatim)

   Copy prompt

   ## Analyze competitor YouTube thumbnails with OpenAI and save approved briefs to Google Sheets
   
   ### Who this is for
   YouTube creators, thumbnail designers, and content teams who want evidence-informed creative directions without copying competitors or pretending that visual analysis guarantees click-through rate.
   
   ### What it does
   This workflow collects a planned-video brief plus two or three public competitor YouTube URLs. It validates and deduplicates the links, retrieves public oEmbed metadata, and sends each public thumbnail to OpenAI for a schema-validated visual analysis. It then aggregates the usable evidence and produces exactly three original thumbnail concepts.
   
   A human review page shows the public sources, the analysis, and all three concepts. Only an approved concept becomes a pre-filled [ThumbCanvas](https://thumbcanvas.com/create-thumbs) link. The approved decision is then appended to Google Sheets as an audit row. Rejected concepts create neither a ThumbCanvas link nor a sheet row.
   
   ### Setup
   1. Add an OpenAI credential to both OpenAI nodes. The default model is `gpt-4.1-mini`; change it in **Set Template Configuration** if your account uses another vision-capable model.
   2. Connect Google Sheets OAuth in **Append Approved Brief to Google Sheets**. Select a spreadsheet and a sheet. Use these header names: `approved_at`, `planned_video_title`, `target_keyword`, `audience`, `competitor_urls`, `approved_concept`, `concept_evidence`, `thumbcanvas_url`, `openai_model`, `analysis_status`, `review_notes`, `workflow_version`.
   3. Publish the workflow, open the production form URL, and submit a brief.
   
   ### Important boundaries
   OpenAI vision can misread small text or visual details, so the workflow exposes sources and limitations for human checking. It never predicts CTR. Enter only public competitor links and non-confidential brief data. OpenAI receives the thumbnail URL and brief context; Google Sheets stores only the approved row. n8n may retain execution data according to your instance settings.
   
   ThumbCanvas generation happens only when the reviewer opens the final link. No image is auto-published.

2. Gather Public Evidence and Vision Analysis

   Each competitor becomes one n8n item. The HTTP Request node calls YouTube’s public oEmbed endpoint for title and author (no YouTube API key required). The OpenAI Vision node analyzes the public thumbnail URL at high image detail and returns a strict JSON Schema covering focal subject, visible text, composition, contrast, emotion, hook, transferable patterns, copy risks, confidence, and limitations. The Validate Structured Visual Analysis node checks schema completeness; failed items are preserved and labeled, aggregating to complete / partial / failed without silently dropping sources.

   Execution prompt · Source text。System Message field of the Analyze Each Thumbnail with OpenAI Vision node

   Copy prompt

   You are a conservative visual analyst. Describe only visible thumbnail evidence. Do not infer views, click-through rate, audience response, or creator intent. Return the required JSON schema.

   Execution prompt · Source text。User Message field of the Analyze Each Thumbnail with OpenAI Vision node (expression)

   Copy prompt

   ={{ 'Analyze this public YouTube thumbnail. Video ID: ' + $json.video_id + '. Public title: ' + $json.source_title + '. Channel: ' + $json.author_name + '. Identify the focal subject, visible text, composition, contrast, emotion, visual hook, transferable design patterns, copy risks, confidence from 0 to 1, and limitations. Small or uncertain text must be marked uncertain.' }}

3. Generate Concepts and Human Review

   A second OpenAI call receives the brief and validated visual evidence, producing exactly three original directions (curiosity gap, high stakes, before and after). Each direction includes source video IDs, rationale, composition, palette, production prompt, and risks. The workflow pauses on an n8n Form review page for up to seven days. The reviewer can open public sources, compare evidence, approve one concept, add notes, or reject all three. Without approval, no ThumbCanvas link is created and nothing is written to Google Sheets.

   Execution prompt · Source text。System Message field of the Generate Three Evidence-Linked Concepts node

   Copy prompt

   You are a YouTube thumbnail art director. Produce original concepts grounded in supplied evidence, avoid copying protected expression, and never claim guaranteed performance. Return the required JSON schema.

   Execution prompt · Source text。User Message field of the Generate Three Evidence-Linked Concepts node (expression referencing the upstream concept_prompt at runtime)

   Copy prompt

   ={{ $json.concept_prompt }}

4. Handoff to ThumbCanvas and Audit

   After human approval, the Resolve Human Approval Code node encodes the chosen production prompt and 16:9 aspect ratio into a ThumbCanvas create URL. Opening that URL is a separate human action and may use the reviewer’s ThumbCanvas plan or credits. The Map Approved Brief to Audit Columns node shapes the result into 12 fixed columns, appended by the Google Sheets node using auto-mapped headers. A final Verify Audit Write Result check distinguishes successful writes from credential or destination failures. If the Sheets write fails, the ThumbCanvas link remains visible but the audit failure is clearly labeled; fix and rerun.

n8n workflow: collect brief → vision analysis → generate concepts → human review → ThumbCanvas handoff and audit