AI thumbnail maker: how to make a thumbnail people actually click

An AI thumbnail maker turns a short description of your video into a finished cover picture: a subject, a background, a mood and, if you want them, a few words of text, in the shape YouTube expects. The fastest way to a good one is to decide the single idea the thumbnail must sell before you write a prompt, generate it wide at 16:9 (or tall at 9:16 for Shorts), make three or four takes, and add or check the words last. This guide walks through that job from start to finish: the sizes YouTube asks for, how to write a thumbnail prompt, how to keep your own face in it, how to get readable text, and when a frame from your own video beats anything generated.

What a thumbnail has to do

A thumbnail is shown small, beside a title, among a dozen others. It gets a moment to say what the video is and why it is worth a click. That sets three rules every good thumbnail follows, whichever tool made it:

  • One idea. One face, one object, one before-and-after. Two competing subjects shrink each other.
  • Readable at the size of a stamp. Big shapes, strong contrast, and at most three or four words. Small detail disappears on a phone.
  • True to the video. A thumbnail that promises something the video does not deliver gets the click once and loses the viewer after that.

Keep those three in mind while you prompt. An AI model will happily fill the frame with detail. Your job is to ask for less.

The sizes YouTube asks for

As of November 2026, YouTube’s own help page on custom thumbnails asks for:

Thumbnail for Aspect ratio Resolution YouTube suggests Minimum
A regular video 16:9 3840 x 2160 640 pixels wide
A Short 9:16 2160 x 3840 640 pixels tall

The page also says thumbnails should be “uploaded in image formats such as JPG or PNG” and stay under its size limits: 2 MB for a video thumbnail uploaded from a phone, and 50 MB from a computer. Custom thumbnails need a verified account. For Shorts, the page says: “Your account must be verified to use this feature.” Check YouTube’s help page before a big upload, because these numbers have changed before.

The practical upshot: generate in the shape you will upload. A square picture cropped to 16:9 loses the top and bottom of the composition, which is usually where the face or the title was.

How to make a thumbnail with AI, step by step

  1. Write the idea in one line. “Me, shocked, holding a tiny camera next to a huge one.” If you cannot say it in one line, the thumbnail will not either.
  2. Pick the shape first. 16:9 for a video, 9:16 for a Short. Then describe the picture for that shape: a wide frame has room for a face on one side and an object on the other, a tall one stacks them.
  3. Describe the subject large. Say how much of the frame it fills: “a close-up face filling the left half of the frame”. Thumbnails work at close-up and medium framing far more often than wide.
  4. Ask for contrast. A bright subject on a dark or plain background, or a saturated colour against a muted one. Name it: “bright yellow jacket against a deep blue background”.
  5. Name the expression or the action. Surprise, focus, laughter, pointing at something, holding it up. A still picture reads better when something is happening.
  6. Keep the background simple. “Plain gradient background” or “blurred studio” beats a detailed room every time at thumbnail size.
  7. Generate three or four takes. The same prompt gives a different picture every run. Pick the one that reads best when you shrink it.
  8. Add or check the words last. Either ask the model for the text (see below) or leave space and add it in your usual editor.
  9. Shrink it before you decide. Look at the take at the size it will appear on a phone. If you cannot tell what it is in a second, simplify the prompt and run it again.

Here is the difference a prompt makes.

A thumbnail about cameras.

That gives you something, but every choice is the model’s.

Wide 16:9 thumbnail. On the left, a close-up of a young man with a shocked expression, mouth open, filling half the frame. On the right, he holds up a tiny compact camera beside a huge cinema camera. Bright, even light on his face, plain deep blue gradient background, saturated colours, sharp focus, photograph. Empty space across the top for a title.

Every phrase in the second prompt removes one guess, and the last sentence leaves room for the words. If writing a prompt like that is the slow part, the free AI image prompt generator turns a rough idea into a full prompt with no account. Our guide to text to image generators covers prompt writing in more depth.

Getting readable text on a thumbnail

Text is where most generated thumbnails go wrong. Some image models are built for lettering and some are not, and the difference shows most on short, bold words.

Two ways that work:

  • Ask for the words in the prompt. Put them in quotes, say where they go and how they look: the words “TINY vs HUGE” in thick white capitals across the top, black outline. Keep it to three or four words. Read every letter of every take, because one wrong letter is enough to make a thumbnail look careless.
  • Leave the space and add the words yourself. Prompt for “empty space across the top” and put the title on in whatever editor you already use. This is slower, and you get exact fonts and spelling every time.

For a channel with a fixed style (the same font, the same colour bar, the same corner logo), the second way is usually the right one. Generate the picture, keep your template for the words.

Keeping your own face in the thumbnail

Most channels put the creator’s face on the thumbnail, and a generated stranger is no use for that. Two approaches:

  • Start from a photo of you. Upload a clear, well-lit photo and use it as a reference picture: “this person, shocked expression, holding a tiny camera, plain blue background”. Models that accept reference pictures keep the face and change the scene. Our guide to making an AI image from an image goes through this in detail.
  • Keep a small set of reference photos. Three or four photos of your face from different angles and in different light help a model hold the likeness across a whole series of thumbnails. The same trick keeps an AI character consistent from shot to shot.

Only use photos of people who agreed to it. A recognisable face of someone else on your thumbnail is a problem whatever tool made it.

A frame from your own video is often the best start

If the video already exists, the strongest thumbnail is often a frame from it. It is honest by construction (the moment really happens in the video), and the face, the light and the set already match.

The way to do this with AI is to use that frame as the starting point. Pause the clip on the strongest moment, save the frame as a picture, then hand it to an image model as a reference with a prompt that cleans it up: “same person and scene, brighter light on the face, plain blurred background, more saturated colour, empty space across the top”. You keep the real moment and lose the clutter around it.

Doing this in Reel Studio

Reel Studio is one way to run the whole job above in one place. It is a studio for making pictures and short clips with the leading models, paid from one credit balance. What it offers for thumbnails, checked against the live site in November 2026:

Step In Reel Studio
The shape Picture models offer 16:9 and 9:16, among other shapes
Choosing a model 15 picture models, including Nano Banana, Seedream, GPT Image, Qwen and FLUX, switched per picture
Text in the picture GPT Image, Qwen and Nano Banana are described in the studio as strong at lettering
Your face Upload your photos to your bank and use them as reference pictures (Nano Banana 2.1 takes up to six)
A frame from a clip Stop any clip on the frame you want and save it to your bank, then use it as a reference
The prompt A prompt assistant rewrites your one-line idea for the model you picked, and you keep only the edits you like
The cost Each picture shows its price in credits on the button before you generate, and a take that fails gives its credits back

To try it, sign in for 40 free credits with no card, which the pricing page puts at about 10 Nano Banana 2.1 pictures. That is enough to make a few takes of one thumbnail and see whether the workflow suits you. After that, plans start at $12 a month for 450 credits, which the pricing page puts at about 112 Nano Banana 2.1 pictures. The same balance makes the video itself, if you make your clips with AI too; our guide on how to make an AI video covers that side.

AI thumbnail maker or a dedicated thumbnail tool?

There are tools built only for YouTube thumbnails, and for some channels they are the better fit. A rough way to choose:

  • A dedicated thumbnail tool suits you if thumbnails are the only picture you make, and you want features built around YouTube specifically, such as templates sized for it or tools that read your video’s title.
  • A general image studio suits you if you also make reel covers, posts, product shots or the clips themselves, and you want to pick a different model when one is better at faces and another at lettering.
  • Your existing editor stays the right place for the words if your channel has a fixed template. Generate the picture, then lay the title on as you always have.

When the manual way is enough

AI is not always the fastest route to a good thumbnail. Do it the usual way when:

  • You already have a great frame. A real reaction, sharp and well lit, needs a crop and a title, nothing more.
  • Your channel has a strict template. If every thumbnail is the same layout with a new photo, a template in your editor is quicker than prompting.
  • The video is a tutorial about a screen. A clean screenshot with a short label often explains a software tutorial better than any generated scene.

Use AI where it saves the slow part: a background you could not shoot, a cleaner version of a cluttered frame, an object or scene that does not exist yet, or a few different ideas to test before you settle on one.

In short

A good AI thumbnail starts with one idea, in the shape you will upload, with a big subject, strong contrast and a simple background. Make several takes, check every letter, and shrink the winner before you trust it. Use your own face or your own frame as a reference whenever the thumbnail should look like you and your video. If you want to try the whole job in one place, Reel Studio gives you 40 free credits to make your first few thumbnails.

← All articles