I Turned He Tongxue's 10 Fake-Bilibili Cover Secrets Into 10 AI Prompts

Updated 2026-04-22 · All 10 rules below are baked into ChatIMG's Video Cover Generator. Skip the reading and go use them.

Why I wrote this

Last week I saw @huangyihe post on X: "ChatGPT helps me optimize covers, not bad. From now on I'm going to do it this way." He attached two covers — original vs. ChatGPT-optimized.

When a marketing KOL makes "ChatGPT for covers" his default workflow, it means: AI cover optimization has gone from novelty to daily-driver.

But here's the problem: generating a pretty cover is easy. Generating a cover that earns clicks is hard. So I dug up a 2023 viral gem from top Chinese creator He Tongxue:

He Tongxue: To Find the Traffic Code, We Built a Fake Bilibili (10M+ views)

He built a fake Bilibili mini-app called "GeliGeli" and loaded 4-6 candidate covers for each of his videos, then let 100+ blind testers vote on which one they'd most want to click. He reverse-engineered "what makes a cover click-worthy" from real user data.

This post turns his experiment's core insights into 10 prompts you can paste directly into any AI generator.

10 Traffic Rules × AI Prompts

1. Simplicity wins: information density is the killer

Insight: A counter-intuitive finding from He Tongxue's experiment — simple versions usually had the highest CTR, no matter how the complex versions varied. Covers stuffed with icons, text, and stickers got lower clicks.

Prompt formula:

A YouTube thumbnail, 1280x720, MAXIMUM 3 visual elements: [main subject],
[one line of bold text], [one accent color]. Rest is solid color or clean gradient.
No stickers, no multiple text layers, no busy patterns.

2. Faces only for veterans: don't show yours if you're new

Insight: Big creator's face = fan trust = traffic code. But a stranger's face is a negative signal — viewers don't auto-trigger curiosity for a face they don't know.

Prompt formula (for newcomers):

A video thumbnail WITHOUT a human face. Use [a strong symbol: tool / product / data / comparison]
as the main subject. Add one bold text overlay in red or yellow.
1280x720, high contrast.

Prompt formula (for established creators):

A video thumbnail, close-up of [specific description: Asian male with glasses,
shocked expression, open mouth] taking 40% of left frame, right side [symbol + text].
Face must be crisp, backlit, eye-catching.

3. Cover text must NOT repeat the title — supplement or gap

Insight: The most common cover text mistake is "write it on the cover, write it in the title" — pure redundancy. He Tongxue's experiment: cover text is only a traffic booster when it adds something the title doesn't say, or creates contrast/suspense.

Prompt formula:

A video thumbnail with ONE line of text. Title is "[your title]".
The cover text MUST NOT repeat the title. Instead:
- Reveal a SURPRISING NUMBER/FACT (e.g. "-47%")
- Pose a QUESTION the title answers
- Create CONTRAST (e.g. "Before → After")
3-6 words max.

4. Color contrast: the "MrBeast red + yellow + black" trio

Insight: High-CTR covers in the experiment share one thing — extreme contrast colors. MrBeast's red + yellow + black strategy follows the same rule.

Prompt formula:

A video thumbnail using only 3 colors: pure red (#FF0000), bright yellow (#FFD700),
and deep black (#000000). Subject in center, red arrow or circle highlighting it,
yellow text overlay. No gradients, no pastels.

5. Exaggerated expressions: smiling isn't enough — go "breakdown/shock/awe"

Insight: The more dramatic the expression, the higher the CTR. MrBeast's "O-mouth", reaction thumbnails — same principle.

Prompt formula:

A video thumbnail featuring a [male/female] face with EXAGGERATED expression:
eyes wide open, mouth in an "O" shape, eyebrows raised to extreme.
Think "reaction thumbnail" style. Close-up, 60% of frame, dramatic lighting.

6. Numbers beat adjectives

Insight: Covers with concrete numbers outperform covers with adjectives ("amazing", "incredible"). Numbers give the brain a verifiable anchor.

Prompt formula:

A video thumbnail with a BIG NUMBER as the focal point: "[specific number]"
displayed in huge bold font (80px+), centered. Supporting visual below.
Numbers like "$10,000", "30 days", "1 million" work best.

7. Arrows and circles: the simplest visual guide

Insight: High-CTR covers use thick arrows, red circles, highlighter strokes to tell the eye "look here in 3 seconds".

Prompt formula:

A video thumbnail with a bright red circle (stroke 8px) around [main subject],
and a bold yellow arrow pointing from text to the circle.
The arrow should be hand-drawn style, not geometric.

8. Before/After is the universal template

Insight: Split-screen Before/After almost never failed in the experiment — Before on the left, After on the right, arrow in the middle. The brain wants to know what happened.

Prompt formula:

A video thumbnail split 50/50: LEFT shows [initial state] in desaturated tones
with text "BEFORE". RIGHT shows [result state] in vibrant colors with text "AFTER".
Arrow pointing right in the middle.

9. "Clean chaos" background: make the subject pop

Insight: Creators stuff backgrounds and bury the subject. Winning covers have extremely simple backgrounds (solid color / soft noise / blurred bokeh) that make the subject feel cut out.

Prompt formula:

A video thumbnail where the subject has strong outline/border against
a MINIMAL background: either pure solid color, or very soft blurred bokeh.
No busy textures behind the main subject. Subject should feel "cut out"
from the background.

10. Content is king. Cover is only an amplifier.

Insight: The experiment's ultimate conclusion — cover+title are amplifiers, not deciders. Bad content + great cover = viewer bounces in 5 seconds. Good content + mediocre cover = CTR climbs over time.

Prompt formula (none — this is mindset):

First ask: can the first 30 seconds of my video hold me? If not, don't spend 3 hours on the cover. Fix the opening first.

5 meta-formulas

After absorbing the 10 rules, here are 5 "meta-formulas" you can rotate endlessly:

[main subject] + [one short line] + [one accent color] — the simplicity formula
[Before → After structure] + [arrow] — the contrast formula
[exaggerated expression] + [shock symbol: ! / ? / burst] — the emotion formula
[big number] + [supporting visual] + [contrast colors] — the data formula
[symbol] (product/tool/icon) + [bold text] — the newcomer formula

Rotate these 5 across your next 10 videos and you're already ahead of 90% of covers.

Verify your instinct against real YouTube data

Methodology is theory. Next step: verification. In our ChatIMG Cover Blind Arena:

Real YouTube thumbnails → view counts / titles / channel names hidden
You guess "which cover earned more views"
Reveal real data, score your cover instinct

After 10 rounds you'll find — the 10 rules above get validated again and again by real-world data.

FAQ

Q: What's the difference vs Canva / Pebblely / Fotor? A: Those are design tools (give you a usable image). We're a traffic tool (give you an image that earns clicks) — the methodology is baked into the prompt, not something you drag-and-drop manually.

Q: Why "rewrite" He Tongxue instead of original research? A: "GeliGeli" is the most rigorous cover A/B experiment by a Chinese creator I've seen — 100+ real users blind-voting — way more credible than any blogger's opinion. We turn his conclusions into actionable prompts and credit his original BV link.

Q: Do these prompts work on English platforms? A: Yes. Colors, contrast, expressions, numbers — these are cross-cultural visual languages. Swap the text to English and go.

Go

→ Enter the Video Cover Generator (10 prompts baked in) → Enter the Blind Arena (real YouTube data) → 30+ open prompt library

Methodology source: He Tongxue: To Find the Traffic Code, We Built a Fake Bilibili (2023-05-16, Bilibili @老师好我叫何同学)