# Make YouTube thumbnails with ChatGPT instead of Canva > Generate on-brand thumbnail backgrounds with ChatGPT image generation, then add your real face and exact headline text with a free HTML template or the OneLessTool in-browser composer. Covers prompts, text-rendering fixes, export at YouTube's sizes and an A/B test before you cancel. - Outcome: I want eye-catching, consistent YouTube thumbnails every week without paying for Canva Pro. - Replaces: Canva (task_specific) - Uses: ChatGPT — Generate thumbnail backgrounds, props and concept variations from a saved series-style prompt; brainstorm 2–4 word headline text.; Google Gemini (optional) — Alternative image generator. Gemini's free tier lists image creation and editing, so it also works if you don't pay for ChatGPT.; Claude (optional) — Alternative for the layout side: ask it to adapt the HTML template (new layouts, colours, fonts). It is better at layout code than at photographic images. - Extra monthly cost: $0 — Assumes you already pay for (or use the free tier of) ChatGPT or Gemini. The HTML template and the OneLessTool composer are free and run in your browser. Image-generation limits vary by plan; heavy users on free tiers may hit daily caps. - Difficulty: easy; skill: basic; setup: 60–90 minutes once (series style prompt, template colours, face cutouts) - Evidence: editorial; illustrative example: true; last reviewed 2026-10-13 - Canonical: https://onelesstool.com/workflows/youtube-thumbnails-with-chatgpt ## Good fit for - Solo creators who pay for Canva Pro mainly or only for thumbnails — for them this can replace the whole subscription - Channels with a repeatable series look (same layout, colours, face position each week) - Creators who already pay for ChatGPT Plus or use Gemini and want one less design subscription ## Not a fit for - Teams where several editors share Canva brand kits, folders and approval comments - Creators who also rely on Canva for social posts, video editing, presentations or printed material (cancelling would remove those too — see the Canva Free downgrade workflow instead) - Thumbnails that need pixel-precise photo compositing of many cutouts; use a real photo editor for that - Anyone uncomfortable with AI-generated imagery on their channel or with disclosure questions it may raise ## Before Open Canva Pro, duplicate last week's thumbnail, swap the background with a stock photo or Canva AI image, remove the background from a new face photo with Canva's background remover, retype the headline using the brand kit font, then download a PNG and upload it to YouTube Studio. ## After Paste a saved series-style prompt plus a short per-video prompt into ChatGPT to get a text-free background, drop it with a transparent face cutout and the exact headline into a reusable HTML template (or the OneLessTool composer), capture a 2560x1440 PNG from Chrome, check legibility at 168x94, and upload. Canva is no longer needed for this job. ## Steps ### 1. Record your baseline before changing anything In **YouTube Studio > Analytics > Content**, note the **impressions click-through rate (CTR)** and views for your last 8–10 videos. Save the numbers in a sheet with the thumbnail file for each video. Also list what you actually do in Canva for thumbnails: background removal, brand fonts, stock photos, templates. Each item needs a replacement in the steps below. If you use Canva for **other** jobs too, this workflow only replaces the thumbnail part. ### 2. Write a series style prompt once and save it Use the **Series style prompt** below to describe your look in words: palette (hex codes), lighting, mood, recurring props, and **which side stays empty for text and your face**. Save it where you can reuse it every week: - ChatGPT: a **Project** with the style text in its instructions, or a pinned note you paste in. - Gemini or Claude: a saved prompt or a doc. The key rule: **ask for backgrounds with no text at all.** Image models still misspell words, and you need exact control over the headline. ### 3. Generate the background for each video Paste the series style plus the **Per-video background prompt**. Ask for a **16:9 landscape** image. Generate 2–3 options and pick the one with the clearest single focal point. - If the model adds stray letters or logos, use the **Text-artefact fix prompt**. - Keep the empty side genuinely empty. Busy areas behind text are the main cause of unreadable thumbnails. - Save the image at the largest size offered. YouTube's help page recommends large thumbnails (it lists 3840x2160 for videos) and a minimum width of 640px. ### 4. Make a real face cutout (don't let AI redraw your face) Take a well-lit photo with the expression you want. Remove the background with a tool you already have, for example the subject-lift feature built into recent macOS and iOS Photos, or your phone's gallery app. Save it as a **transparent PNG**. You *can* ask ChatGPT or Gemini to cut out or "enhance" your face, but generated edits can subtly change your features. Viewers notice. Use a real photo for your face and AI only for the scene. Build a small library of 6–10 cutouts (surprised, pointing, thinking, laughing) so most weeks need no new photo. ### 5. Compose the thumbnail with exact text Two options, both free: 1. **OneLessTool thumbnail composer** on this workflow page: drop in the background and face PNG, type the headline, pick the accent colour and export. Everything runs in your browser. 2. **The HTML template below**: save it as `thumbnail.html`, replace the `{{placeholders}}` (image paths can be local files next to the HTML, e.g. `bg.png`), and open it in Chrome. The page shows live previews at 168x94, 320x180 and 480x270 under the main canvas. Keep the headline to **2–4 words**. Use the **Headline options prompt** to get choices that add to the video title rather than repeat it. Keep the bottom-right corner clear: YouTube overlays the video length there (the template marks this area in red). ### 6. Export at the right size and check the file In Chrome with the template open (add `?export` to the URL to hide guides): 1. Open DevTools and toggle the **device toolbar**. 2. Set the size to **1280 x 720** and the **device pixel ratio to 2** (gives 2560x1440) or **3** (gives 3840x2160). 3. In the Elements panel, right-click `
` and choose **Capture node screenshot**. Then check the file: - Width at least **1280px**. YouTube's A/B test tool downscales *all* variants to 854x480 if any variant is below 1280x720. - **Under 2 MB** if you upload from the YouTube mobile app (desktop allows up to 50 MB). On a Mac, `sips -s format jpeg -s formatOptions 85 thumb.png --out thumb.jpg` converts a large PNG to a smaller JPG. - JPG or PNG format. Custom thumbnails require a **verified** YouTube account. ### 7. Test against your Canva thumbnails before cancelling For the next 4–6 uploads, use YouTube Studio's **A/B testing (Test & compare)** on desktop: upload your new thumbnail alongside a Canva-made version (up to 3 variants). YouTube picks the winner by **watch time**, not CTR, so also compare CTR against the baseline from step 1. Cancel Canva Pro only when the new thumbnails win or tie on most tests and your time per thumbnail is acceptable. Download anything you need from Canva before the subscription ends (see checklist). ## Prompts ### Series style prompt (save and reuse) Define a consistent visual identity for every thumbnail background in your series. ```text You are generating background images for YouTube thumbnails for a channel about {{channel_niche}}. Apply this series style to every image I ask for in this conversation: - Format: 16:9 landscape, photographic, high contrast, sharp focus on one subject. - Palette: dominant {{primary_color_hex}}, accent {{accent_color_hex}}, deep shadows. Avoid muddy mid-tones. - Mood and lighting: {{mood}}; strong rim light on the main subject. - Recurring elements: {{recurring_elements}}. - Composition: put the main subject on the {{subject_side}} third. Keep the {{empty_side}} 40% of the frame simple and darker, with no objects, so I can add a headline and a person there later. - Never include any text, letters, numbers, logos, watermarks or UI elements. - No people unless I ask. Never generate a face meant to be me. Reply "Style saved" and wait for my first video brief. ``` ### Per-video background prompt Generate one background per video using the saved series style. ```text New video brief, same series style as before. Video topic: {{video_topic}} The single visual idea: {{key_object_or_scene}} Emotion the viewer should feel at a glance: {{emotion}} Leave the {{headline_side}} side empty and darker for text. Make 1 image. Then suggest 2 alternative scene ideas in one line each, so I can ask for them if this one doesn't work. ``` ### Text-artefact fix prompt Remove stray or misspelled text from a generated image, or fix text when you decide to keep it in the image. ```text Edit the last image. Keep the composition, subject, lighting and colours exactly the same. Remove every letter, number, sign, logo and watermark, including any on {{object_with_text}}. Replace those areas with plain matching texture. If I give exact text below, render it instead, spelled exactly, in capitals, in a heavy {{font_style}} style, positioned {{text_position}}, and nothing else: Exact text (may be blank): "{{exact_text}}" After editing, list any characters you were unsure about. ``` ### Headline options prompt Get short thumbnail text that complements, not repeats, the video title. ```text Video title: {{video_title}} What happens in the video / the hook: {{video_hook}} Audience: {{audience}} Write 10 thumbnail headline options of 2–4 words each. Rules: do not repeat words from the title; create curiosity or state a concrete result; no clickbait claims the video doesn't deliver; avoid words longer than 9 letters. Mark your top 3 and explain each in one short line. ``` ### Small-size critique prompt Get a second opinion on legibility and focus before uploading (attach the exported image). ```text I've attached a YouTube thumbnail. Imagine it shown at about 168x94 pixels next to {{competitor_count}} other thumbnails in the {{channel_niche}} niche. Answer briefly: 1) Can the headline "{{headline_text}}" be read at that size? 2) What is the first thing the eye lands on? 3) Is anything important in the bottom-right corner, where YouTube shows the video length? 4) Two specific changes that would most improve it. ``` ## Templates and scripts ### HTML/CSS thumbnail template (1280x720 layout with live small-size previews) (template) ```html Thumbnail – {{headline_line_1}}

{{headline_line_1}}{{headline_line_2}}

{{badge_text}}
168 x 94 (small/mobile check)
320 x 180
480 x 270
``` ## Limitations - Image models still misspell text and add stray letters; exact headlines need to be added separately (template or composer). - No Canva-style drag-and-drop editing of the generated image; changes mean re-prompting or editing the template. - Generated images vary between runs, so a strict series look takes a saved style prompt and some re-rolls. - Background removal for your face depends on another free tool (OS photo apps); results on hair and glasses can be rough. - Image-generation limits depend on your ChatGPT or Gemini plan and can change; free tiers may cap daily images. - No shared brand kit, folders or approvals for a team. ## Risks - **ip_licensing** (medium): Don't prompt for copyrighted characters, logos or a real person's likeness. Check the image tool's terms on commercial use of outputs; YouTube thumbnails must also follow Community Guidelines. - **functionality_loss** (low): Cancelling Canva Pro removes Pro features for every other design you make there, not just thumbnails. - **data_portability** (low): Old Canva thumbnails stay editable only inside Canva. Download final PNGs of past thumbnails you may want to rework before cancelling. - **reliability** (low): Thumbnail quality may drop for a few weeks while you learn the prompts; the A/B test step guards against this. ## Verification checklist (run before cancelling) - [ ] You have produced at least 3 thumbnails end-to-end with this workflow, each in under 30 minutes. - [ ] Each exported file is at least 1280px wide (ideally 2560 or 3840), 16:9, JPG or PNG, and under 2 MB if you upload from mobile. - [ ] At the 168x94 preview in the template, someone else can read the headline and name the subject within 2 seconds. - [ ] Nothing important sits in the bottom-right corner covered by the video-length badge. - [ ] Your face cutout is a real photo, with clean edges on hair and shoulders at full size. - [ ] YouTube Studio A/B tests on 4–6 videos show the new thumbnails winning or tying against Canva versions, and CTR is not below your recorded baseline. - [ ] Any Canva designs or assets you still need are downloaded before the subscription ends. ## Savings methodology Savings = what you currently pay for Canva Pro (or your share of a Canva plan) − incremental cost of this workflow ($0 if you already have ChatGPT or use Gemini's free tier). Count the full Canva price only if thumbnails are the only reason you pay; if you use Canva Pro for other work, the saving is $0 until those jobs are replaced too. One-time setup (about 1–1.5 hours) and any extra minutes per thumbnail are not monetised. Example (illustrative): Solo creator posting one video a week, paying for Canva Pro monthly only for thumbnails, already paying for ChatGPT for scripting. $15/mo → $0/mo. Assumptions: Illustrative: the $15/month Canva Pro figure is a round placeholder. canva.com blocked automated access on 2026-10-13, so we could not verify the current US price; check your own billing page. ChatGPT is already paid for other work, so its cost is not counted. Thumbnails are the only reason the creator pays for Canva Pro. ## Sources - YouTube Help: Add video thumbnails: https://support.google.com/youtube/answer/72431 (accessed 2026-10-13) - YouTube Help: A/B test titles & thumbnails: https://support.google.com/youtube/answer/13861714 (accessed 2026-10-13) - Google Gemini subscriptions: https://gemini.google/subscriptions/ (accessed 2026-10-13) - Canva pricing: https://www.canva.com/pricing/ (accessed 2026-10-13)