How to make a thumbnail from your own photos that gets clicked

Last updated 2026-09-01 · 8 min read

What is the short version?

A thumbnail is a face, a few words and a colour. The face is the part people get wrong: they pick the photo where they look best, when the one that works is the photo where they look like the video's subject.

1. Go to Vizard Agent and send several photos with different expressions.
2. Give the title of the video.
3. Give one accent colour and let it lead.

What do you need before you start?

Four or five photographs of yourself, and the video's actual title. Vizard Agent chooses between them by expression rather than by composition, so send genuinely different ones — shocked, delighted, thoroughly unimpressed — rather than four versions of the same pleasant smile.

• The photos. Several, with real expression range.
• The title. The words the thumbnail has to support.
• The accent colour. One hex value.
• The subject. What the video is actually about.
• The platform. It sets the size and the safe area.

What do you type into Vizard Agent?

Give it the choice to make rather than handing over the answer. Vizard Agent picks whichever expression is strongest for the title you supplied, and that is reliably a better decision than the one most people make when judging photographs of their own face.

Prompt

Variants worth knowing:

• Send a range. Different expressions, not different angles.
• One accent colour. Two makes it noisy at small size.
• Give the real title. The image has to argue with it.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real thumbnail. The first two steps are the ones that decide the result, and the middle third is entirely about getting a cutout that survives a dark background.

1. Downloads and tiles the input photos to inspect them together.
2. Inspects the facial expressions across all of them.
3. Checks the background-removal options available.
4. Uploads the chosen pose for cutting out.
5. Checks the tools for background generation and for the cutout.
6. Checks which image-processing libraries are available.
7. Tests local subject segmentation on the images.
8. Installs a dedicated background-removal library when the first routes fall short.
9. Removes the background from the selected pose.
10. Extracts the transparent cutout and its alpha channel.
11. Checks the cutout quality over a test background and reviews it.
12. Generates a high-contrast background with accent lighting.
13. Reviews that background, then generates a second concept and reviews it too.
14. Checks the available bold typefaces for the title.

Step eleven is the one people skip. A cutout looks perfect against white and falls apart against a dark background, where every stray pixel around the hair becomes a visible halo — so Vizard Agent composites it over a test ground before building anything on it.

What does the result look like?

The thumbnail this page is written from was built from four supplied photos: the one with the strongest expression cut out cleanly, set against a generated high-contrast background with accent lighting, with the title in a genuinely bold typeface.

Two different background concepts were generated and compared before one was chosen. That is worth doing — the background is where a thumbnail either says what the video is about or says nothing at all.

When does this not work well?

A thumbnail gets judged at roughly the size of a postage stamp, on a phone, in a list of twenty other thumbnails. Almost every way one fails is a direct consequence of that, rather than of anything you would notice looking at it full size.

• Low-resolution photos. The cutout edge shows.
• Hair against a busy background. The hardest cutout there is.
• Long titles. Four or five words is the practical limit.
• Two accent colours. They fight at small size.
• Flattering over expressive. The nicer photo usually loses.

How do you fix a result that came back wrong?

Say which of the three elements is wrong — the cutout, the background or the type. Vizard Agent keeps the transparent cutout, both generated background concepts and all the type settings, so changing one of them does not disturb the other two.

• "There is a halo round my hair." Cutout refined and re-checked.
• "Use the other photo." Swapped, and the layout re-balanced.
• "The text is unreadable small." Fewer words, set larger.

How does Vizard Agent compare to doing it yourself?

By hand this is a background removal that looks fine until you place it, a background hunted for online, and a title that turns out to be three words too long. Vizard Agent tests the cutout over a real background and generates the ground to suit the subject.

Common questions

How many photos should I send? Four or five with genuinely different expressions. Vizard Agent chooses between them.

Why not the most flattering one? Because a thumbnail is an argument, not a portrait. The expression has to match what the title claims.

Will the cutout be clean? It checks. Vizard Agent composites it over a test background and refines the edge if it shows.

Can it generate the background? Yes, and usually should. Vizard Agent builds one that says something about the subject.

How many words in the title? Four or five. Vizard Agent will say when the text will not read at thumbnail size.

Can it match my channel's colours? Yes. Vizard Agent takes a hex value and builds the accent lighting around it.

Will it look like my other thumbnails? If you want. Vizard Agent can work to an existing set's treatment.

What size does it output? Your platform's. Vizard Agent renders large and checks it at small size.

Can it make several options? Yes. Vizard Agent generates different backgrounds around the same cutout.

Does it work without a face? Yes, though faces perform better. Vizard Agent builds around a product or an object instead.

Why check the cutout on a dark background? Because that is where it fails. Edges that look perfect on white show a bright halo the moment they sit on something dark, and a thumbnail's background is rarely white.

Can I reuse the cutout? Yes. Vizard Agent keeps it as a transparent file you can reuse for future thumbnails without cutting yourself out again.

Should the thumbnail repeat the title? No — it should add to it. Vizard Agent writes the few words on the image to say something the title does not already say.

Does it check the result? Yes. Vizard Agent reviews the cutout, both background concepts and the finished layout before delivering it.

What if none of my photos work? It will say so rather than picking the least bad. Vizard Agent tells you what expression the title needs so you can shoot one.