Thumbnails that get clicked: my method after years on YouTube

The thumbnail matters more than the video. That is harsh and true: an excellent video with a bad thumbnail is not watched, and an average video with a good one is.

I used to lose a lot of time on them. Enough time that I seriously considered hiring a graphic designer.

Then I stopped.

The trick that changed everything#

The big problem with generating an image containing your face is that the face is not you. It comes out close, but it is not you, and viewers feel it even when they cannot say why.

The fix is annoyingly simple: I photographed my face once, from every angle. Front, profile, from above, different expressions. A small library of my own.

And I now send them with every prompt.

Since then the problem has gone. The face is mine because it is actually mine, and the tool handles everything else.

Why I reject a thumbnail#

After hundreds of attempts, my reasons are four and they do not change:

Clutter. Too many things in frame. A thumbnail is seen at thumb size on a phone, and what is not understood in half a second is not understood.

Unreadable text. Either too small, or over a busy background, or in a typeface that does not survive being shrunk. In Arabic it is harder, because Arabic type loses legibility faster.

Tool errors. Six fingers. A hand at an impossible angle. Text that resembles letters and does not read. These are more than half of what I throw away.

A face that is not mine. Solved by the trick above, but before it, this was the number one cause.

What I actually do#

A face, always. With a specific expression, not a generic smile. A face showing something definite, surprise or doubt or concentration, outperforms a neutral one.

Three words maximum. The image says half and the title says the other half. Repeating the title inside the image wastes the space.

High contrast. Because most viewers see it in a crowded list.

The thumb test. Shrink it to the size of your thumbnail and look. If you cannot tell what it is about, throw it away. That test alone improved my results more than anything else.

Arabic specifically#

Two points that never appear in translated articles:

Direction. The Arabic eye starts from the right. Put the text on the right and the face on the left, not the reverse. The difference is real.

Typeface. Thin Arabic faces vanish when shrunk. Use a heavy weight, heavier than you think you need.

The surprises#

And I will be honest: my predictions are wrong constantly.

Thumbnails I thought were weak performed well, and ones I laboured over went unnoticed. Which is why I stopped trusting my instinct alone and started testing.

That is the real lesson in all of this: nobody knows what works before publishing it. Anyone selling you a guaranteed formula is selling an illusion.

In closing#

Photograph your face once, send it with every prompt, throw away half of what comes back, and test at thumb size. That is my entire method, and there is no deeper secret in it.

Try this today: take your last published thumbnail and shrink it to fingernail size. If you cannot tell what the video is about, you have found a reason your clicks are low.

Common questions#

How do I get my face right in generated images?

Photograph your face once from every angle with different expressions, then send those photos with every prompt. The more real material you give the tool, the less it invents.

How many words should a thumbnail contain?

Three at most. The image carries half the idea and the title carries the other half; repeating the title inside the image wastes the space.

What is different about thumbnails for an Arabic audience?

The eye starts from the right, so put text right and face left, and use a heavy Arabic typeface because thin ones disappear when shrunk.

What is the fastest quality test?

Shrink it to fingernail size and look. If you cannot understand the subject in half a second, discard it however good it looks at full size.