Carousel, Thumbnail & Visual Design
What this skill is
A working method for designing social visuals that are readable at a glance on a phone: carousels, thumbnails, prompt cards, and pin visuals. It covers the brief, the layout, the type scale, the pre-publish gate, and the weekly research habit that keeps the work fresh.
Who it is for: anyone who briefs or checks visuals for social posts — a designer, a creator, or an agent writing visual briefs for an image-generation step. This skill writes briefs and checks layouts; it does not generate images. You supply your own image tool.
Prerequisites
- An image-generation or design tool (any tool that takes a brief and returns an image).
- A way to view images at full size and at phone (stamp) size.
- A contrast checker (any WCAG contrast tool) for the text/background pairs you use.
- A facts source for your post: every number or claim that appears on a visual must come from it (see "Canonical-facts lock" below).
The design system (example — substitute your own)
The palette and type choices below are one worked example system, a dark editorial look. They are the default in the examples only. Pick your own palette and fonts; keep the structure (one dark ground, one light text colour, one accent) and keep the contrast rule.
- Example palette: near-black
#0A0A0C(ground) · warm yellow#FFC933(accent) · cream#F5F2EA(main text) · one restrained grey for secondary text. - Example type: one condensed display face for headlines, one clean sans for body. Never more than two typefaces on one visual.
- Per-post theme varies inside the system: same colour family, different layout, accent balance, imagery, and composition per post. Two near-identical visuals in one day is a defect — a viewer scrolling should be able to tell them apart at a glance.
Hard rules
- Never big text + empty background + button. Every visual carries hook + visual story
- information + proof + hierarchy + brand. A headline floating on a flat field with a button is not a design.
- Real interfaces only. Show the actual product, tool, or screen the post is about. No fake app screenshots, no generic robot stock art, no random neon effects, no decorative motion without meaning.
- Thumbnail promise matches the content. A thumbnail that promises something the content does not deliver is a critical failure. Never use a deceptive thumbnail.
- Contrast: text meets WCAG 2.1 Level AA — at least 4.5:1 for normal text and 3:1 for large text (Success Criterion 1.4.3). Check every text/background pair; do not eyeball it.
- Safe zones respected. Keep all text well inside the frame on every side. Text that touches, crosses, or runs off any edge is a defect — rebuild, never ship.
- Alt text for every meaningful image. Write one plain sentence describing what the image shows and what it means.
- No unreadable tiny text. If a word cannot be read at phone size, cut it or make it bigger. Never shrink a prompt or a block of text to squeeze it beside other content.
- Text is built in luma, not hue, at small sizes. On 4:2:0 video and compressed images, colour-only edges smear. Small text needs a strong brightness difference from its background, not just a different colour.
- Every visual gets a full text scan before it ships — read every word on the image against the brief, one by one. Any typo, wrong word, duplicated word, or cut-off text means the visual is corrected (regenerated or rebuilt) before the post goes out.
Dimensions (exact — check before export)
| Format | Size (px) | Notes |
|---|---|---|
| Carousel (portrait) | 1080 x 1350 | 7–10 slides; slide 2 is the second hook |
| Pin | 1000 x 1500 | Exact 2:3 — do not ship another ratio |
| Shorts / story cover | 1080 x 1920 | 9:16 |
| Standalone feed visual | 1080 x 1350 | Same as carousel slide |
| Document carousel / PDF first page | 1080 x 1350 | First page carries the hook |
The 15-point visual gate (hard gate)
One FAIL = revise, never ship. Check every visual against all 15 before it goes out.
- Hook at a glance. A viewer understands the point in under 2 seconds.
- Hook + visual story + information + proof + hierarchy + brand are all present — not just a headline on a flat field.
- Real subject. The visual shows the real product, tool, screen, or result — no fake screenshots, no generic stock art.
- Thumbnail promise matches the content. Nothing promised that the post does not deliver.
- Contrast passes WCAG AA (4.5:1 normal text, 3:1 large text), checked with a tool.
- Safe zones. All text well inside the frame on every side.
- Full text scan passed. Every word on the image read against the brief; no typo, wrong word, duplicate, or cut-off text.
- Numbers and claims match the facts lock. Every figure on the visual is identical to the entry in your canonical-facts file (see below).
- No ghost words. No faint, half-formed, or unreadable pseudo-text anywhere on the image (see "Ghost-word discipline").
- Slide 2 is a second hook (carousels). The carousel does not go quiet after the cover.
- Ranking cards carry all six elements (when the post ranks things): ranking badge · name · "best for" · key benefit · limitation · where-to-find-it cue.
- Prompt cards are readable. A prompt appears as a clean, readable code-style block or prompt card, at a size a phone reader can actually read. Never shrunk to fit.
- Alt text written. One plain sentence per meaningful image.
- Dimensions exact. The exported file matches the size in the table above.
- Theme is distinct from the last visual. Same system, different layout, accent balance, and composition — not a near-copy of yesterday's post.
Record one gate line per visual (e.g. VISUAL-GATE: 15/15 PASS) in the file where you
keep the post copy.
Visual brief format
Every brief carries all eight fields. A brief missing one is incomplete — do not send it to the image step.
- Theme name — unique per post, one short phrase.
- Layout sketch — slide-by-slide for carousels (see patterns below), or a single layout description for a standalone visual.
- Palette balance — which colour dominates, which accents, and where.
- Headline text — 3–8 words, hook at a glance, written verbatim.
- Proof element — the real UI, screenshot, chart, or result the visual shows.
- Text density check — word count per slide and the smallest text size used.
- Dimensions — from the table above.
- Alt text — the sentence you will publish with the image.
A filled-in example brief is in examples/visual-brief-example.md.
Slide-by-slide layout patterns
Use these as starting patterns, not templates to copy blindly. Every post still gets its own composition inside the pattern.
Pattern A — Listicle / ranking (e.g. "5 tools that do X")
- Slide 1 (cover): headline (3–8 words) + the count ("5 tools") + one strong focal image. No list on the cover.
- Slide 2 (second hook): the payoff — what the reader gets by swiping ("the one most people skip is #4"). This slide decides whether the swipe happens.
- Slides 3–7 (one item each): ranking badge (large, top corner) · name · "best for" (one line) · key benefit (one line) · limitation (one line, honest) · where-to-find-it cue. Same structure, different accent balance and imagery per slide.
- Last slide: recap in one line + the single next action. No new information here.
Pattern B — Tutorial / how-to (e.g. "Do X in 4 steps")
- Slide 1 (cover): outcome headline ("Do X in 4 steps") + the finished result shown, not described.
- Slide 2 (second hook): the hardest part named honestly ("step 2 is where most people stall") + what makes it doable.
- Slides 3–6 (one step each): step number (large) · action in one line · the real screen or result for that step · one common mistake to avoid.
- Last slide: the full sequence in miniature (4 steps, one line each) + next action.
Pattern C — News / alert (e.g. "X just changed — what it means")
- Slide 1 (cover): the change in plain words + "NEW" or date marker + one focal image. No hype words the content cannot support.
- Slide 2 (second hook): who it affects and who it does not — the eligibility line, on the slide, not buried later.
- Slides 3–5: what changed (real screenshot or quote) · what it means in practice · what to do now (one concrete action).
- Last slide: the one-line summary a reader can repeat to someone else.
Pattern D — Build-along (e.g. "Build X with me")
- Slide 1 (cover): the thing being built, shown finished + the time or effort it takes.
- Slide 2 (second hook): the starting point — what you need before step 1.
- Slides 3–8 (one stage each): stage name · the real screen or artefact at that stage · one decision the reader must make at that stage.
- Last slide: the finished artefact + what to build next.
Thumbnail concept-to-layout spec
A thumbnail is briefed in three stages. Never go from a one-line idea straight to a final.
- Brief (one paragraph). The video/post title, the one feeling the thumbnail should create (curiosity, surprise, clarity), the focal image, and the exact cover text (verbatim, 1–4 words that ADD to the title instead of repeating it).
- Three concepts. Three distinct layouts from the same brief — different focal image, different text position, different accent. Each concept is one sentence: what is in the picture, where the text sits, what feeling it creates.
- Chosen layout. Pick one concept. Write the full layout spec: focal image, text (verbatim), text position and size, background, accent, safe margins on all four sides. Then make it. Check the finished thumbnail at full size AND at stamp size (the size it appears in a feed). If the text is unreadable at stamp size, or touches any edge, or any word is misspelled, regenerate — never ship it.
Thumbnail lab workflow (weekly, 30 minutes)
Run once a week, same day each week.
- Pick the week's most important post and write its thumbnail brief.
- Generate the three concepts. Compare them side by side at stamp size, next to the thumbnails they will appear beside in a feed.
- Pick the winner by one question: which one gets clicked first, and why? Write the answer in one line.
- Make the one change that would make the winner win by more. Regenerate once if needed.
- File the brief, the three concepts, and the winner's one-line reason with the post.
Type scale + spacing tokens
Recommended starting values. They are design tokens, not laws — adjust them to your fonts, but keep the ratios and the minimum sizes.
| Token | 1080 x 1350 | 1000 x 1500 | 1080 x 1920 |
|---|---|---|---|
| Headline | 110–140 px, bold | 100–130 px, bold | 120–150 px, bold |
| Sub-headline | 56–64 px | 52–60 px | 60–72 px |
| Body | 40–46 px | 38–44 px | 42–48 px |
| Caption / label | 30–34 px | 28–32 px | 32–36 px |
| Smallest text allowed | 28 px | 26 px | 30 px |
| Outer margin (all sides) | 72 px | 64 px | 84 px |
| Gutter between elements | 36 px | 32 px | 40 px |
| Line height (headline) | 1.05x | 1.05x | 1.05x |
| Line height (body) | 1.35x | 1.35x | 1.35x |
Rules for the tokens:
- Headline 3–8 words. More words means smaller type; smaller type means a weaker hook. Cut words before you cut size.
- One headline size per visual. A second, slightly smaller headline competes with the first and both lose.
- Minimum sizes are floors. Text below the floor is unreadable at phone size — cut the words instead.
- Margins are safe zones. Nothing — text, logo, or badge — crosses the outer margin.
Ghost-word discipline
Ghost words are words an image generator renders that look like text at a glance but are faint, half-formed, misspelled, or unreadable on close inspection — letter-like shapes in a background, a blurred label, a sign that says nothing. They are a defect, not a detail, because a viewer reads them as sloppy or deceptive even when they cannot make out the letters.
- Scan every generated image at full size specifically for ghost words before the general text scan. Zoom into backgrounds, signs, screens, and book spines.
- Any ghost word = regenerate or rebuild. Never crop around it and hope nobody notices.
- If a generator keeps producing ghost words in the same spot, change the brief: remove the background element that carries the fake text, or replace it with a plain field.
- Real text that is intentionally small (a genuine UI label) is not a ghost word — but it must still be spelled correctly and readable at phone size, or it is cut.
Canonical-facts lock
A canonical-facts lock is a single file you maintain listing every fact and figure your visuals are allowed to state: product names, prices, dates, counts, rankings, and quotes — each with its source. It is the only source of truth for numbers on visuals.
- Every number, price, date, and quoted claim on a visual must match the lock exactly. If the lock says "37", the visual says "37" — never "40" or "37+".
- A figure whose qualifier matters ("over", "more than", "estimated") carries the qualifier ON the visual, next to the figure, at headline size — never only in the caption or the voiceover.
- If a fact is not in the lock, it does not go on a visual. Either verify it and add it to the lock first, or cut it.
- When a fact changes, update the lock first, then rebuild every visual that states it. Never patch the visual and leave the lock stale.
Weekly creative-research routine (30–45 minutes)
Run once a week. The point is principles, not copying.
- Sample (10 minutes). Collect 10 high-performing visuals in your niche from the past week — feeds, search results, or a platform's trending surface. Save them with their source and date. Do not download or reuse anyone's image; you are studying layout, not taking assets.
- Record (10 minutes). For each one, write one line: what the hook is, where the text sits, what the focal image is, and what makes it stop the scroll.
- Extract principles (10 minutes). Look across the 10. Write 3–5 principles in your own words (e.g. "the winning covers put the result, not the tool, in the top half"). A principle is a rule you can apply to a different topic — never a layout to copy.
- Test one variable (next post). Pick one principle and apply it to one visual in the coming week. Change one variable only, so you can tell whether it helped. Note the result next to the principle.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Visual looks like every other post | Same layout + same accent balance repeated | Change the layout pattern and the accent balance, not just the headline |
| Text unreadable at phone size | Below the minimum token, or colour-only contrast | Raise the size to the token floor; check WCAG contrast; use a brightness difference, not just hue |
| Generator keeps misspelling a word | The word is long or the brief buries it | Shorten the headline to 3–8 words; put the exact text in quotes in the brief; regenerate |
| Ghost words in backgrounds | Generator filling empty zones with fake text | Remove the background element from the brief; replace with a plain field |
| Thumbnail loses at stamp size | Text too small or focal image too busy | One focal image, 1–4 words, high contrast; test at stamp size before shipping |
| Numbers on the visual are challenged | Visual disagrees with the facts lock | Fix the lock first if the lock is wrong; rebuild the visual from the lock |
Known limits
- This skill designs and checks visuals; it does not generate them and it does not guarantee performance. A visual that passes the gate can still underperform — the weekly research routine is how you learn what your audience actually responds to.
- Contrast ratios are for text legibility. They do not make a bad layout good.
- Dimensions and platform behaviour change. Re-check the size table against the platform's current specs before a format you have not shipped in a while.
- The example palette and type scale are starting points. If your brand system differs, substitute it and keep the gate — the gate is the skill, the palette is an example.