Prompting for expressions: what each of the eight faces says

An expression is a claim about the size and direction of what happens in the video. Choose the rung, not the most dramatic face.

By the Thumbnail Bench team Published 8 min read
A matrix of six video types against a first-choice expression, a second choice and the expression to avoid.
On this page
  1. The eight faces sit on three ladders
  2. What each face says
  3. The pairs creators confuse
  4. Choosing by video type
  5. Where the eyes go
  6. Describing an expression in a prompt
  7. Changing the expression on a finished cover
  8. What to do next

An expression on a thumbnail is a claim about the size and direction of what happens in the video, so the right one is chosen by the outcome, not by which face looks most dramatic. Face Lab produces eight expressions from one selfie (Shocked, Surprised, Happy, Excited, Angry, Curious, Scream and Sad), and they are not eight equal options. They sit on three ladders, each rung making a bigger or a different claim than the one below it, and the most common mistake is picking the right ladder and climbing one rung too high.

The eight faces sit on three ladders

Our view

This is our framing rather than anything measured, but it has held up across the covers we look at. Four of the eight expressions are one emotion at rising intensity: Curious, Surprised, Shocked, Scream. Two are a positive emotion at two intensities: Happy, Excited. The last two are negatives that differ in where they point: Angry says someone is at fault, Sad says nobody is. Seen this way, choosing an expression becomes two decisions: which ladder, then which rung.

A horizontal scale from calm on the left to overwhelming on the right, with markers for Curious, Surprised, Shocked and Scream.
The surprise ladder. Each rung claims a bigger outcome, and a rung above the video's real outcome is a promise the video cannot keep.

The rung matters because the viewer reads it as the size of the outcome. A Shocked face promises something big; if the video's twist is modest, the viewer who clicked for big leaves early, and YouTube's help page on A/B testing titles and thumbnails says a test's result is decided by watch time share, not clicks. Promise alignment for AI covers in general is covered by the guide to how generators work and where they fail; here the practical rule is to pick the rung that matches the outcome and, when unsure, the rung below.

What each face says

The table describes what each expression does physically, because that is what a viewer reads at thumb size, and what claim it makes about the video.

ExpressionWhat the face doesWhat the viewer readsFits
ShockedBrows high, eyes wide, mouth openSomething big and unexpected happenedReveals, results, fails, big numbers
SurprisedBrows up, eyes wide, mouth closed or barely openI did not expect thisDiscoveries, small twists, comparisons, tests
HappyRelaxed smile, soft eyesThis went well and I like itTutorials, positive reviews, day-in-the-life
ExcitedOpen smile, brows up, visible energyI cannot wait to show you thisUnboxings, launches, wins, announcements
AngryBrows down and in, mouth tight or shoutingSomeone did something wrongWarnings, rants, exposés, complaints
CuriousOne brow up, slight squint, head tiltedIs that true?Questions, myths, tests, finance and education
ScreamMouth wide, face tense, eyes wide or shutOverwhelmingGaming, horror, extreme challenges
SadBrows drawn in, mouth down, eyes loweredI lost somethingConfessions, failures, goodbyes, mistakes

A spec card listing the eight expressions with what the brows, eyes and mouth are doing in each.
At thumb size the brows and eyes carry the expression; the mouth mostly sets the intensity.

The general question of which emotion fits which promise, and why the face is read before anything else, belongs to the guide to facial expressions on covers. This piece assumes that and goes one level down, to the specific eight.

The pairs creators confuse

Most expression mistakes are not a wrong ladder but a wrong rung, and four pairs account for nearly all of them.

PairThe tellChoose the first whenChoose the second when
Surprised vs ShockedThe mouth: closed or barely open against wide openThe twist is real but modest, or the audience distrusts dramaThe outcome is large and the niche accepts energy
Happy vs ExcitedState against anticipation: a settled smile against an open oneThe video is the pleasant thing itselfThe video is about the thing that is coming
Curious vs SurprisedBefore against after: a question against an answerThe video asks and testsThe video found something
Angry vs ScreamDirection against volume: pointed at someone against pointed at nothingThere is a culprit in the videoThere is only intensity
Observed pattern

Surprised is the most under-used face in the set. Creators reach for Shocked because it is louder, and in finance, education and reviews the louder face reads as hype to exactly the viewers who were going to click. The closed-mouth version carries the same "I did not expect this" with credibility intact, which is why it appears in the first-choice column for more video types than any other expression in the hero figure.

Take a cover for a video where a £30 charity-shop lamp turns out to be worth £400. The instinct is Scream: mouth wide, hands on the head, the lamp small in a corner, INSANE FIND in red. The viewer infers a stunt channel and discounts the number before reading it. One rung down is Surprised: lips together, eyes wide, the lamp large and lit at its edge, the face turned toward it, £400? as the only words. Same outcome. The face now looks like someone who was not expecting it rather than someone paid to react.

We would render Surprised over Shocked on any reveal that is not a stunt, even though the Shocked version will sometimes win the first glance. The click it wins is the one that leaves early, and the test resolves on watch time share, not on the glance.

Choosing by video type

Work from the outcome. Name what the video delivers in one sentence, decide whether that is a question, an answer, a win, a loss or a wrong, then pick the ladder and the rung.

The video isFirst choiceSecond choiceAvoid
A reveal or a resultSurprisedShockedHappy (it answers the question)
A review or a verdictCuriousAngry if the verdict is bad, Happy if goodScream
A lesson or a methodHappyCuriousShocked
A challenge or a stuntScreamExcitedSad
A confession or a storySadSurprisedExcited
A warning or an exposéAngryCuriousHappy
A versus or a comparisonCuriousSurprisedScream

Two niche notes, both observed patterns: finance audiences tend to trust Curious and Surprised and read Shocked as a sales pitch, which is the reasoning behind the doubt-led finance cover; education audiences tend to reward Happy and Curious because both signal "this is manageable", the case made in the calm educational cover.

Where the eyes go

The expression sets the claim; the gaze sets the eye path. Surprised and Curious work best looking at the object, because they invite the viewer to look at what caused them. Angry and Sad work best looking at the viewer, because they are addressed to someone. Shocked and Scream can do either, and Happy and Excited looking at the viewer read as an invitation. The mechanics of gaze, and when eye contact reads as sales, are the subject of eye direction and gaze in thumbnails. The rule to remember here is that the face should be looking at whatever it is reacting to, and that thing should be in the frame or in the headline.

Describing an expression in a prompt

When the cover has no saved face, the expression has to be written into the scene prompt, and the single emotion word is the weakest way to do it. "Surprised" gives the model a wide target. The mechanical description gives it a narrow one.

ExpressionPrompt wording that lands
Shockedeyebrows raised high, eyes wide open, mouth open, leaning back slightly
Surprisedeyebrows raised, eyes wide, lips together, head still
Happyrelaxed genuine smile, eyes slightly narrowed, shoulders down
Excitedwide open smile, eyebrows up, leaning toward the camera
Angryeyebrows pulled down and together, jaw set, lips pressed or mouth shouting
Curiousone eyebrow raised, slight squint, head tilted, mouth closed
Screammouth wide open, face tense, eyes wide, neck strained
Sadeyebrows drawn in and up at the centre, corners of the mouth down, eyes lowered
Thumbnail Bench recommends

Write one expression per prompt and describe brows, eyes and mouth in that order, because that is the order of importance at thumb size. With a saved face, choose the expression in Face Lab and keep the prompt about the scene; asking the prompt for a different emotion than the face you selected gives the model two instructions for the same region. The six-part prompt structure puts the emotion in part one for the same reason: it narrows everything after it.

Changing the expression on a finished cover

A render is often right in every way except the rung: the scene, the light, the headline and the composition all work, and the face is Shocked where the outcome was Surprised. Rendering again risks all of it. On the bench you can change the expression on a finished result instead, which returns two options for two credits and leaves the rest of the picture alone.

That is also the cleanest one-variable change for a test. Two covers that differ only in expression, Curious against Surprised on the same scene and headline, ask one question of the audience, which is the discipline set out in designing test variants that resolve.

All eight expressions come from one selfie, so the quality of that photo decides the quality of every face after it; what makes a selfie work in Face Lab covers the framing, light and neutral base. And the face has to be one you may use: your own, or a co-host's or client's who agreed to be on your thumbnails. No public figures without consent, no minors, as the acceptable use policy sets out.

What to do next

Write the one-sentence outcome of your next video. Decide whether it is a question, an answer, a win, a loss or a wrong; that names the ladder. Then choose the rung one step below your first instinct and render two versions, the instinct and the step below, from the same scene. Judge them at thumb width beside the covers already in your niche's row, and keep a note of which rung this audience trusted, because that note is worth more than the credit it cost.

Thumbnail Bench team

We build an AI thumbnail maker and spend our days looking at what earns clicks in the feed. Thumbnail Bench was created and is run by Tim Schroeder, a YouTube creator of several years; the founder notes are his. Everything here is written by the team, checked against YouTube's own documentation where a claim can be checked, and labelled as observation or opinion where it cannot. About us.

Ready to go viral?
Make your first thumb.

4 free credits, 4 more after your first render. No card. Your first options in about a minute.