YouTube Thumbnails With AI: Types, CTR and How to Make Them
YouTube thumbnail types that work, the exact 1280x720 spec, and how to make A/B test variants with AI in minutes on VdoBloom.
A YouTube thumbnail is 1280x720 pixels, and the types that work fall into a handful of recognisable layouts: a face with big stacked text, a pointing pose with a highlighted keyword, a small figure in a scene under one huge headline, or a split screen of the topic and your reaction. VdoBloom’s Thumbnail Maker generates three or four of those layouts at once from one photo of your face and a typed title, ready to load into YouTube’s Test & Compare.
Updated October 2026.
Disclosure: VdoBloom is our platform.
What a thumbnail actually has to do
A thumbnail has one job: persuade someone scanning a crowded feed to click, and then not disappoint them when they do. That second half matters more than most people think. YouTube’s own thumbnail testing tool picks its winner by watch time share rather than raw clicks, which means a thumbnail that wins clicks but loses viewers is not a success.
The design consequences are consistent. A thumbnail is viewed small, often on a phone, so one subject and one readable phrase beat a busy composition. Contrast does the heavy lifting, because the image sits against a white or dark feed depending on the viewer’s theme. And the bottom-right corner should be left clear, because that is where the video’s length badge sits.
Thumbnail types, and when each one works
These are the recurring formats rather than categories YouTube defines. The layout descriptions match the four concepts VdoBloom’s tool produces, which is a useful short list to design against.
| Type | Layout | Works well for |
|---|---|---|
| Face plus stacked title | Face on the right, two or three words stacked on the left | Commentary, reactions, opinion and vlogs where you are the subject |
| Pointing with a keyword | A pointing pose, an arrow and one highlighted word | Tutorials, explainers and list videos where attention needs directing |
| Small figure, huge headline | A small figure in a scene with one very large headline across the top | Story-driven videos and topics where the setting is the point |
| Split screen | The topic on one side, your reaction on the other, title band at the bottom | Reviews, comparisons and anything with a surprise or reveal |
| Before and after | Two states side by side, usually with a divider | Transformations, repairs, makeovers and process videos |
| Object on plain background | One product or object, centred, clean background | Product reviews, unboxings and gear videos |
| Number or count | A large numeral as the dominant element | Listicles and ranked content, where the count is the promise |
Whichever you pick, the same rule applies: everything the thumbnail promises has to actually be in the video. YouTube’s spam policy bans maliciously misleading thumbnails — imagery that tricks someone into clicking a video that does not deliver what was shown. An explosion, a celebrity or a result the video never contains is a policy problem, not just a style choice.
Thumbnail specifications
| Property | Value |
|---|---|
| Recommended size | 1280x720 pixels |
| Aspect ratio | 16:9 |
| Minimum width | 640 pixels |
| Maximum file size | 2 MB |
| Accepted formats | JPG, PNG, GIF and BMP |
| Shorts cover | A different asset entirely — see the Shorts aspect ratio guide |
Because both the thumbnail and the channel banner are 16:9, one design can be reworked into the other rather than redrawn, though the banner needs its text inside a much smaller safe area — the banner size guide has those numbers.
How to make thumbnails with AI on VdoBloom
- Upload a face photo, or pick a frame from your video. A clear, well-lit shot works best. Your face, skin tone and features are kept from the photo and only the expression changes, so it stays recognisably you.
- Type the title and the topic. The title field holds up to 60 characters and is reproduced character for character, so it prints exactly as you type it. The topic field takes up to 160 characters and describes what the video is about, which is what drives the scene behind you.
- Choose a style. There are four: Bold & loud, with a saturated background, thick white title letters with a black outline and a hard rim light; Clean, with one solid bright background, bold sans-serif title text and soft even light; Dramatic, with a dark moody background, one strong colour accent and glowing title letters; and Playful, with pastel pop colours, chunky rounded letters and doodle accents.
- Choose how many variants. Three is the right number for YouTube’s Test & Compare, which accepts up to three. Pick four if you want a spare to swap in later.
- Generate and download all. The variants come back side by side with a single download button. Each one uses a different layout from the table above, so the test has genuinely different concepts to measure rather than three variations of one idea.
- Load them into Test & Compare and let the data choose.
One detail the tool handles for you: it keeps the bottom-right corner of the artwork free of text, because the video length badge covers it. The tool also upscales the output, and defaults to 2K.
What the variants cost
Each image costs the same as a single image on the model you pick, so a set of three costs three times the per-image price. On GPT Image 2.5 Flare, the cheaper of the two models the thumbnail maker uses, that is 5 credits an image at 2K — so 15 credits for three variants, or 20 for four. The other option is Nano Banana Pro at 18 credits an image at 2K. The exact total is shown before you generate, and failed generations are refunded automatically. Full rates are on the pricing page.
Getting the text right
The title field is the part most people get wrong, and it is worth knowing why the limit is 60 characters: the text is printed exactly as typed, so a long title simply becomes small text. A thumbnail reads at a glance or not at all. Aim for two to four words that carry the promise of the video, and let the topic field do the explaining.
Keep the sentence case simple, avoid punctuation that competes with the image, and check the contrast against the background the tool has generated — a clean bright background needs dark text, or the outline the Bold & loud style adds. And never put a number in the thumbnail that the video does not deliver.
A/B testing properly
YouTube Studio’s Test & Compare accepts up to three thumbnails per video and shows them evenly to viewers, picking a winner by watch time share rather than clicks. Results can take a few days to two weeks depending on impressions and how different the variants are, and the outcome is one of Winner, Preferred or None.
The last outcome is the one to avoid by design. Three near-identical thumbnails usually come back as None, because there is nothing meaningful to measure. Test genuinely different concepts — a different layout, a different expression, different text placement, a different colour scheme — which is exactly what generating three layouts at once gives you.
Before any of that, though, it is worth being clear on the policy question, because it comes up constantly: YouTube treats creating a thumbnail with generative AI as production assistance that needs no disclosure label. What it does police is misleading imagery and the Community Guidelines. Our post on whether YouTube penalizes AI thumbnails covers the rules in detail, and if the video itself contains realistic AI scenes of real people or events, that is a separate disclosure setting at upload — see AI video and YouTube monetization for how that affects earnings.
Common mistakes
- Testing three variations of one idea. Test three concepts instead, or the test returns no clear result.
- Crowding the frame. One subject, one phrase, strong contrast. Anything else is invisible on a phone.
- Letting text fall into the bottom-right corner, where the length badge sits.
- Using a face that is not in the video, or a real public figure who has nothing to do with it. That edges into impersonation and misleading-clickbait territory.
- Forgetting the Shorts cover, which is a 9:16 asset and not the same file at all.
Frequently asked questions
What size should a YouTube thumbnail be?
1280x720 pixels, which is 16:9, and no larger than 2 MB. The minimum accepted width is 640 pixels. Because the thumbnail is 16:9, a design made for it can be reworked into a channel banner rather than drawn again.
What are the main types of YouTube thumbnail?
The recurring layouts are a face with stacked title text, a pointing pose with a highlighted keyword, a small figure in a scene under one huge headline, a split screen of topic and reaction, a before-and-after pair, an object on a plain background, and a large number for list content. VdoBloom’s thumbnail maker produces the first four as separate variants from a single photo.
Does YouTube penalize AI-made thumbnails?
No. YouTube’s disclosure guidance lists using generative AI to create a thumbnail as production assistance that does not need a label. What applies to every thumbnail is the rule against misleading clickbait, plus the normal Community Guidelines.
How many thumbnails can I test at once?
Up to three per video with YouTube Studio’s Test & Compare, which picks the winner by watch time share rather than raw clicks. VdoBloom’s thumbnail maker generates three layouts by default, or four if you want a spare.
How much does an AI thumbnail cost?
One image costs what the chosen model charges per image. On GPT Image 2.5 Flare that is 5 credits at 2K, so three variants cost 15 credits and four cost 20. On Nano Banana Pro it is 18 credits an image. The exact total shows before you generate, and failed generations are refunded.
Upload one photo of your face, type the title and the topic, and generate three different thumbnail layouts in a single run from the YouTube Thumbnail Maker. New accounts start with 10 free credits and no card, and paid output is watermark-free with commercial rights.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →