Skip to main content

How to make a YouTube thumbnail with AI

Updated

Make a YouTube thumbnail with AI: decide the idea, write a prompt that works, get your face in, fix the text and avoid the generic AI look. With examples.

Quick answer

Decide the idea first: one subject, one emotion and two to four words. Then write a prompt that names the subject, the expression, a plain background, the composition with text space on one side, the lighting and the style. Use reference photos for your face, change one thing per attempt, and check the result at 168 pixels wide.

Key takeaways

  • The AI renders the idea; it does not choose it. Write down the subject, the emotion and the two to four words before you write the prompt.
  • A good prompt names seven things: subject, expression, background, composition, lighting and color, style, and text.
  • Your face comes from reference photos, not from a description. Long text and logos are better added by hand afterwards.
  • Change one thing per attempt, fix small details by editing that area, and check every version at 168 pixels wide.
1234
  • 1Expressive face
  • 2Bold 3-5 word text
  • 3High-contrast color
  • 4One clear focal point
Anatomy of a thumbnail that gets clicks

An AI image generator can give you a lit, composed scene in under a minute that would take hours to shoot or build by hand. It can also give you a waxy face, garbled letters and six fingers. The difference is mostly the input. This guide covers the method that works with any AI image tool: what to decide before you type, how to write the prompt, how to handle your face and the text, and how to check the result before it goes live.

What AI does well for thumbnails, and what it does badly

Good atWeak at
Complete scenes: a kitchen on fire, a desert at dusk, a cluttered gaming deskYour exact likeness without reference photos
Dramatic lighting: rim light, neon, golden hourMore than four or five words of text
Composition on request: subject left, empty space rightBrand logos and real product packaging
Shots you could never take: a phone splitting in half, a car on the moonSmall precise details: dials, screens, labels
Speed: several directions in the time one manual mockup takesHands and fingers, especially holding objects

Want a thumbnail like this in 30 seconds?

Describe your video and let the AI design it. No Photoshop, no design skills.

Generate my thumbnail free

Plan around the right column. If the thumbnail depends on a readable product label or a specific logo, generate the scene and place the real image or logo yourself in an editor.

Decide the one idea before you write anything

Write three things on paper: the subject (one person or one object), the emotion or state (shocked, proud, burnt, broken, finished) and the two to four words. For a video about fixing a slow laptop that might be: a laptop with a glowing fan, a relieved face, FROM 9 MIN TO 20 SEC. If you cannot fill all three, no prompt will rescue it. The AI renders an idea; it does not pick one for you.

Anatomy of a thumbnail prompt

ElementWhat to writeExample
SubjectOne person or one object, described physicallya man in his thirties with a short beard, grey hoodie
Action or expressionOne clear expression or actionmouth open, eyebrows raised, holding a cracked phone
Setting or backgroundOne simple place, or a plain colorplain dark blue background with a soft gradient
CompositionWhere the subject sits and where the text goessubject in the left third, empty space on the right for text
Lighting and colorLight direction and two or three colorsstrong orange rim light from the right, high contrast
StylePhotographic or illustrated, with camera termsphotograph, 50mm lens, shallow depth of field
TextThe exact words in quotes, or no textbold white text reading "IT BROKE"

Three full example prompts

  • Cooking: Close up photograph of a burnt, smoking lasagna in a glass dish on a dark stove, a woman in the left third with both hands on her cheeks, eyes wide. Warm kitchen light from above, plain dark background, shallow depth of field. Empty space top right for text. Bold yellow text reading "I RUINED IT".
  • Personal finance: Photograph of a young man in a white t-shirt holding a thick stack of receipts, frowning at them, positioned in the left third. Plain light grey background, soft window light from the left, natural skin texture, 85mm lens. Right half empty for text. No text.
  • Gaming: Illustrated scene of a knight in dented armor standing in front of a giant red dragon, seen from a low angle. The knight is small on the left, the dragon fills the right two thirds. Dark purple sky, orange fire light on the armor, high contrast. No text.

Example

Weak: "Exciting thumbnail about saving money, colorful, viral." Improved: "Photograph of a woman in her twenties holding a glass jar full of coins, smiling with her mouth open, in the left third. Plain mint green background, soft front light, natural skin texture. Empty space on the right. Bold black text reading "$500 SAVED"." The weak prompt describes a mood; the improved one describes an image.

Getting your own face in

Describing yourself in words gets you a stranger who shares your hair color. If you appear on the channel, use a tool that takes reference photos of your face: several clear, well lit shots from slightly different angles, no sunglasses, no beauty filters. In the prompt, describe only the expression and the pose, and let the photos carry the likeness. Viewers recognize a regular face in the feed, so a lookalike costs you more than no face at all.

Text: ask for it, or add it yourself

Current generators handle two to four short words in capitals reasonably often, and fail more with every word you add, with long words and with anything outside English. Put the exact words in quotes in the prompt. If the letters come out wrong after two tries, stop retrying: generate the image with no text and the empty space you asked for, then type the words in any editor in a heavy font with a thick outline. That takes two minutes and is always correct.

Avoiding the generic AI look

  • Ask for one subject. Crowds, extra props and busy backdrops are where the plastic look shows first.
  • Write natural skin texture, visible pores, or a real camera and lens (35mm, 85mm). It pulls the image toward a photograph and away from a render.
  • Prefer a plain or blurred background over a detailed fantasy setting unless the video is fantasy.
  • Cut words like epic, viral, hyper detailed and 8k. They push every image toward the same glossy style.
  • When one detail is wrong, a hand or an object, edit that area instead of regenerating the whole image. A new generation changes everything you liked too.
AI-powered

Turn this idea into a real thumbnail, right now

  • 9 click-ready designs in one go
  • No design skills required
  • Tuned to lift your click-through rate
Generate my thumbnail free

Free to start. No credit card. Cancel anytime.

Iterate one change at a time

If the second attempt changes the expression, the background and the lighting at once, you will not know which change helped. Keep the prompt fixed and edit one element per attempt: first the expression, then the background color, then the light. Save the prompt that worked, because it becomes the template for your next video and keeps your thumbnails consistent.

Check it before you upload

Shrink the image to 168 pixels wide, the size of the desktop suggested column, and look at it next to other thumbnails. The subject and the words should still read. View it on a white and on a dark background, since viewers use both modes. Leave the bottom right corner clear for the video length. Export JPG or PNG at 1280 x 720 or larger. If you have two strong versions, YouTube Studio's Test & compare can test up to three thumbnails on a computer and picks the winner by watch time share.

Warning

AI makes it easy to show things that never happened: a bigger explosion, a product that is not in the video, a reaction you never had. YouTube's thumbnails policy does not allow a thumbnail that misleads viewers into thinking they will see something that is not in the video. Show a scene the video delivers.

Doing it in ThumBEAST

  1. Sign in with a free account (no credit card) and open the "Create with AI Prompt" tab on the generate screen.
  2. Type your description. If you only have a short idea, "Enhance prompt" rewrites it into a detailed description for free.
  3. Next to the prompt, optionally pick a template, a face you have added (upload three well lit photos once and it is saved for later thumbnails), objects and a background.
  4. Choose the text: type up to 6 words, let the AI decide, or no text.
  5. Press "Generate". You get one 16:9 thumbnail per generation in about 30 seconds.
  6. Open it in the "Image Editor" to select an area and describe what to change, use "Remove BG", or draw arrows, shapes and text. You can also ask for changes in the chat.

Downloads on the free tier carry a ThumBEAST watermark; any paid plan removes it. If the video is already uploaded as public or unlisted, the "Video to Thumbnail" tab is the other route: paste the link, ThumBEAST reads the transcript and builds the thumbnail from what the video says, so you skip writing the prompt.

Common mistakes

  • Opening the generator before deciding the subject, the emotion and the words.
  • Prompts made of adjectives (epic, colorful, viral) instead of a described image.
  • Asking for a full title as thumbnail text and accepting misspelled letters.
  • Regenerating from scratch to fix one hand.
  • Not asking for text space, then covering the subject's face with the words.
  • Showing a scene, an object or a reaction the video does not contain.

Frequently asked questions

Is it allowed to use AI thumbnails on YouTube?

YouTube's thumbnails policy is about what the image shows, not the tool that made it. It does not allow thumbnails that mislead viewers about what is in the video, among other content. An AI thumbnail that represents the video honestly is treated like any other.

Can I make AI thumbnails for free?

Many tools have a free tier. In ThumBEAST you need a free account with no credit card, new accounts get a few free thumbnails, and free downloads carry a watermark that any paid plan removes.

Can AI put my face on a thumbnail?

Yes, if the tool accepts reference photos of your face. Upload clear, well lit photos from a few angles without sunglasses or filters, and describe only the expression in the prompt.

Can AI write the text on a thumbnail?

Usually for two to four short words in capitals, less reliably for longer phrases or other languages. If the letters come out wrong, generate without text and add the words in an editor.

How long should a thumbnail prompt be?

Around 30 to 60 words is enough to cover subject, expression, background, composition, lighting, style and text. Longer prompts tend to add clutter rather than control.

Why do my AI thumbnails look fake?

Usually too many elements, glossy style words and no camera terms. Ask for one subject, natural skin texture, a real lens and a plain background, and remove words like epic and 8k.

What size should an AI thumbnail be?

16:9, at least 1280 x 720, as JPG or PNG. YouTube recommends 3840 x 2160 and accepts up to 2 MB from the mobile app and 50 MB from a computer.

AI-powered

Your next viral thumbnail is 30 seconds away

ThumBEAST's AI designs scroll-stopping YouTube thumbnails for you. Describe your video and get options you can publish today.

Generate my thumbnail free

Free to start. No credit card. Cancel anytime.