TutorialsHands-on: One Story, Three Images
Using a community reading event as the through-line, three prompts in sequence — portrait poster, landscape cover and slide visual, all produced live, each with the full copy-ready text.
Case: the complete promotional set for a community reading event on 26 October.
One poster to share in the group, one article cover for the post, one slide visual for the wrap-up deck — the same event in three different placements, and therefore three different ratios.
All three prompts are copy-ready; replace the bracketed details with your own. The case brief (event name, date, venue, theme colour, exact words) is downloadable under "Course materials" on the overview page.
Three images, three ratios
| Purpose | Ratio | Key constraint | Who sees it |
|---|---|---|---|
| Event poster | Portrait 3:4 | Title in the upper third, details below, no more than 5 text levels | Someone scrolling the group chat, deciding in 2 seconds |
| Article cover | Landscape 16:9 | Text area on the left; headline still readable on a phone thumbnail | Someone scrolling the article list |
| Slide visual | Landscape 16:9 | No text, no real data or names, use metaphor | A room looking at a projected deck |
Run A: event poster (portrait 3:4)
Copy-ready · portrait 3:4
帮我设计一张活动宣传海报。
主体：社区读书分享会宣传海报
场景：温暖的社区书屋，木质书架，暖黄色灯光，几个人围坐读书
风格：扁平插画风格，红色与米色为主色调，喜庆温暖，适合社区活动
构图：竖版3:4，主标题在上方三分之一处，信息区在下方
文字：主标题"读书分享会"，副标题"让阅读点亮生活"，信息小字"10月26日 14:00 · 社区活动中心"
约束：画面干净不杂乱，文字清晰可读，不超过5个信息层级
开始前，先问我三个问题，确认信息后再开始Demo point: run the same prompt twice
This prompt will be run twice and you will see two different images — that is re-rolling, not failure.
- Unhappy? Re-roll and pick the best of the batch;
- Judge on three things only: are the words right, is the image clean, does it feel like the intended tone;
- Remember the principle: AI produces, you select, fix the text and sign off — you are the client.
Two details worth noticing in this prompt
Once the poster is out, read every word on it. Date, time, venue, event name — AI makes mistakes with full confidence and will never point them out.
"10月26日" rendered as "10月16日" happens constantly, and it is always the group chat that spots it, after publication.
Run B: article cover (landscape 16:9)
The cover decides the click-through rate. On a phone the image is tiny — big headline, generous whitespace, or nobody taps.
Copy-ready · landscape 16:9
帮我设计一张公众号文章封面图。
主体：一篇讲"AI帮普通人省时间"的文章封面
场景：一个人轻松喝咖啡，旁边悬浮着完成的文档、日程表图标
风格：现代扁平风，红色渐变背景，商务又不严肃
构图：横版16:9，左侧留出文字区，视觉重心在右
文字：大字标题"AI帮你省时间"，小字"3个方法 · 今晚8点见"
约束：文字大而清晰，手机小图也能看清标题
开始前，先问我三个问题，确认信息后再开始Three things that must be in the prompt
| Point | How to write it | Why |
|---|---|---|
| 16:9 is a hard ratio | Say "landscape 16:9" in composition | Without it, the image gets cropped at both sides when placed |
| Leave the text area empty | "Text area on the left, visual weight on the right" | Artwork overlapping the headline makes it unreadable |
| Put the use case into the constraint | "Large, clear text; headline readable on a phone thumbnail" | How you will use it is the best constraint you can write |
The check: shrink it to thumbnail size
After generating, shrink the image very small and look at the headline.
- Still readable: the cover works;
- Not readable: the headline is too small or the frame is too busy. Go back and harden the constraint — for example "headline takes one third of the image height" — and run again.
In class we run it a second time and compare the two side by side. Comparing makes the difference obvious in a way that looking at one image never does.
Run C: slide visual (two steps)
A slide visual does not need to be beautiful; it needs to be on topic. That is why it takes one extra step: translate the theme into visual language first, then generate.
Step one: let AI write the prompt
Describe the need to a chat assistant and ask it to write the prompt using the five parts. Do not describe the picture here — give the theme, the purpose and the limits, and let AI imagine the visuals.
Step two: take it to the image tool
Paste the prompt AI wrote into the image tool, landscape 16:9, generate twice and pick one for the deck.
Step one · what you send to AI (copy-ready)
我要做一页PPT，主题是"团队本月用AI提效50%"，需要一张配图。
请先问我三个问题（PPT风格、汇报对象、画面比例），
然后帮我写一段即梦AI的文生图提示词。
要求：不出现真实数据与人名，用隐喻（如齿轮加速、火箭升空），16:9横版，风格与商务PPT协调。Step two · the prompt AI returns (paste into the image tool)
主体：一组加速运转的齿轮与向上箭头，象征团队提效
场景：简洁商务空间，浅蓝渐变背景，不出现人物
风格：扁平商务插画，蓝白主色，干净利落
构图：横版16:9，主体居中，右侧留出文字区
约束：图上不放文字，不放真实数据与真实人名Three rules for slide visuals
- No real data and no real names — use metaphor instead (accelerating gears, a rising rocket, puzzle pieces closing);
- Lock the ratio to 16:9 to match your deck, or you get black bars above and below;
- Match the style of your deck template — do not drop a warm orange illustration into a cool-toned deck.
The common trap: asking for "an image about efficiency".
Too abstract, so AI guesses — it may draw someone holding a rocket, or a table with flames, and neither fits your deck.
The right move is to let AI translate "efficiency" into visual language first (accelerating gears, a rising curve, closing puzzle pieces), pick the most on-topic one, then generate.
Not happy? Change one thing at a time
Some part of every image will be wrong. Do not rewrite the prompt — use the editing phrasing below, and change only one thing:
这张图保留整体构图和配色，只修改：【如"主标题改为XX""把背景换成夜晚""人物换成3人"】。
其他元素不变。The rule: one change at a time. Say too much and AI changes the parts you liked — ask it to "make the background night, change to 3 people, and retitle it to 全民阅读月" and you may lose the font, the layout and everything else.
| What you want to fix | What to say |
|---|---|
| The headline is wrong | Keep the composition and palette, change only the headline to "全民阅读月", leave everything else unchanged |
| The background is too bright | Keep the composition and text, change only the background to night, leave everything else unchanged |
| Too many or too few people | Keep the composition and text, change only to 3 people, leave everything else unchanged |
Key points from this lesson
| Point | In one line |
|---|---|
| One story throughout | Same event, three purposes — you practise changing the ratio, not the content |
| Re-rolling is normal | Same prompt, different images; pick the best of the batch |
| Never skip constraints | Without them AI fills the frame |
| Lock 16:9 | Covers and deck visuals are landscape; omit the ratio and they get cropped |
| Thumbnail check | Readable at thumbnail size and the cover works |
| Two steps for deck visuals | Let AI translate the theme into visuals before generating |
| One change at a time | Say too much and AI changes what you liked |
Understand First: Four Terms and the Five-Part Formula
Learn the four concepts, match yourself to six scenarios, then memorise the heart of the course — subject, scene, style, composition, text. Miss one and the image drifts.
Take It Home: Eight Templates and Five Red Lines
Eight prompt templates ordered by your working week, a seven-row visual brief sheet, a tool-selection table organised by scenario, and five red lines to check before publishing.