# Thumbnail system — the visible benefit, the hidden bill

## The repeatable format

A proposed high-curiosity format, **not measured high CTR**. One everyday benefit occupies roughly the left two thirds. One unmistakable cost or contradiction occupies the right third. Join them through gaze, object alignment or a receipt—not red arrows and a cluttered collage. At most one face, one primary object and one consequence. Make the emotional question understandable before any text is read.

Examples: a hand buys a glowing-free, ordinary clock face with a missing 11-minute wedge while an older person's phone shows a missed call; a perfectly repaired bicycle opposite a courier's scraped helmet; a blank ID sleeve opposite the brother who still physically exists. These are concept sketches to commission, not generated images or claims about actual footage. Do not show events the episode does not deliver.

## Visual specification

- Compose at 1280×720,16:9 as a planning master; verify current upload requirements before publishing.
- Grounded editorial illustration; no source-frame references or copied face designs. Cobalt carries the benefit; yellow identifies a specific invoice/choice; cloud/ink keep silhouettes clear. Use mint sparingly for genuine cooperation rather than a second unrelated brand.
- Reserve bottom-right approximately 20% width ×18% height for platform duration overlays; important proof stays elsewhere. This is a conservative composition choice, not a guaranteed platform safe area.
- Build two meaningful focal masses, not many tiny panels. A split is a relationship, not a requirement to draw a divider.
- Optional overlay **0–3 words**, large, plain, manually typeset. Never paste the full title onto the image. Prefer `WHO PAID?`, `NOT MY YES`, `49 MINUTES` only when supported.
- Title names the transaction; thumbnail reveals the unsettling discrepancy. Don't let both repeat the same sentence.
- Check at160×90 and320×180, in grayscale, light/dark surroundings and with duration overlay. Crop shouldn't cut off the cost. Avoid microprint as essential evidence.

## First five launch pairings

| Episode | Title | Thumbnail A | Thumbnail B | Optional words |
|---|---|---|---|---|
|1|I Rented an Extra Hour. Someone Else Lost It.|Mara holding clock receipt; Nell's unanswered phone|Hospital doorway behind Mara; receipt joins a second hand|WHO PAID?|
|2|My Brother Sold the Memory of Signing My Name.|Ivo recognizes his handwriting but not the form|Empty memory slot beside signed family line|MY NAME?|
|3|My Bad Luck Started Happening to One Stranger.|Pristine workshop wheel vs Eli's scraped helmet|Untouched workshop beneath clean umbrella; wet courier outside|HIS BAD LUCK|
|4|The Company Came to Collect Seven Nights of Sleep.|Nell's seven checked nights beside a collection slip|Clinician reading donor line across from alert Nell|SEVEN NIGHTS|
|5|Someone Used My Voice to Say Yes.|Mara pointing to written NO beside playback YES|Mara's hand at throat; counterparty holding voice receipt|NOT MY YES|

Each remaining episode has a specific thumbnail brief inside its story file. Keep the pattern recognizable without making every image another hand holding a receipt. Identity, absence, service, queue, and public choice can become the contradiction.

## Testing without inventing results

1. Before launch: show A/B at small size to a few target readers in randomized order. Ask what happened, what question they'd expect answered, and whether the image/title misled them after reading. Preference is qualitative, **not CTR**.
2. At actual YouTube publication: use platform-native thumbnail testing if this account has it. Compare genuinely different hypotheses, not only colors. Freeze unrelated title changes during the comparison where possible.
3. Record eligible impressions, traffic source, CTR, watch time per impression if available, opening retention, and test dates. Do not compare a homepage-heavy upload to a search-heavy upload as equivalent.
4. A higher click rate with worse satisfaction is not a win. Prefer a truthful version that brings viewers who stay. Small samples and sequential swaps are inconclusive; do not declare a universal benchmark or guarantee.
5. Avoid fabricated timers, deaths, money amounts, fake platform UI, celebrity likeness, fake true-story claims and spoilers that promise a scene absent from the episode.

## Reusable original illustration prompt

Original editorial-realism thumbnail for clearly fictional illustrated story [episode premise]. Left: [one concrete benefit]. Right: [one concrete supported human cost]. One clear visual relationship [gaze/object alignment/negative-space absence]. Stable character bible: [approved original descriptors]. Bright readable scene, flat controlled cobalt/ink/cloud palette with one yellow evidence accent. Strong silhouette and shape separation at tiny size, calm background, maximum one face and two principal objects. Leave bottom-right free of essential evidence and [location] free for optional later typesetting. No text, numbers, logos, watermarks, arrows, gradients, glow, source frames, copied creator layouts, horror clichés, or imagery absent from this story. 16:9.

Prompt is not a paid-generation instruction. Capability, rights, originality review and explicit approved spend remain required. Test a manually composed rough first.