TutorialsUnderstand 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.
No theory in this lesson. It does one job: understand the terms, then memorise the formula.
You cannot place an order without knowing the menu. Every prompt that follows is essentially filling these five slots.
Four terms, plainly
| Concept | In plain words | When you use it |
|---|---|---|
| Text to image | You describe a picture in words and AI draws it. The more precise the words, the closer the result | Building an image from scratch |
| Prompt | The description you feed the AI. Writing one is placing an order with a painter: what to draw, in what style, at what size, with what words | Every single generation |
| Reference image | You hand AI an image and ask it to adapt — new style, new background, new product | You have an image but need it changed |
| Text rendering | AI's ability to write words inside the image. It used to produce gibberish; now Chinese titles usually come out right | Any asset that carries a headline |
Once these four terms are clear, everything else is filling them in. Today's focus is text rendering — because it is the line between "pretty but useless" and "ready to publish".
Six scenarios — which one is yours
| Scenario | The trade name | What AI does |
|---|---|---|
| Event poster | Key visual — the face of an event | Produces the scene and writes the title, date and venue on it |
| Article cover | Header image — the first gate to click-through | 16:9 covers with a large title, three minutes each |
| Slide / deck visual | Illustration or mood image — so the deck is not all text | Concept images, process diagrams, data-visualisation backgrounds |
| WeChat Moments / community image | Promo asset — it has to earn the tap | Portrait images with headline, selling point and QR space |
| Knowledge card | Quote card or note card — the most shareable format | A consistent template producing a whole series at once |
| Store / product image | Product or lifestyle shot — looks sell | Reference-image redraw: new background, new season, one style |
You do not need all six. Pick the one you actually have to deliver next week, get its template right, and copy the pattern for the rest.
One formula: the five parts
Subject ＋ scene ＋ style ＋ composition ＋ text — this is the heart of the course. Miss one and the image drifts.
| Slot | The question it answers | Example (community reading event) |
|---|---|---|
| Subject | What is being drawn | A poster for a reading event |
| Scene | What environment and background | Inside a warm community library, wooden shelves, warm yellow light, a few people reading together |
| Style | What mood and tone | Flat illustration, red and beige, warm and festive |
| Composition | What layout and ratio | Portrait 3:4, title in the upper third, details below |
| Text | What words appear | Title "读书分享会", small text "10月26日 14:00" |
Join those five sentences and you have a working prompt:
A poster for a community reading event. Scene: warm community library, wooden shelves, warm yellow light, a few people reading together; Style: flat illustration, red and beige, warm and festive; Composition: portrait 3:4, title in the upper third, details below; Text: title "读书分享会" plus date and venue in small type.
Each slot explained
The slot people forget is "text". Write "a poster for a reading event" and AI returns a lovely blank illustration. Write "Text: title '读书分享会', small text '10月26日 14:00 · 社区活动中心'" and it actually puts the words on the image.
Cannot remember five items? Remember one line: what to draw, where, in what mood, how big, with what words — it is exactly how you would brief a designer.
One extra line doubles your success rate
Add this to the end of every prompt:
"Before you start, ask me three questions and confirm the details first."
This turns a one-way instruction into a two-way confirmation: AI asks who the event is for, how many text levels you want, whether there is a fixed brand colour — and fills the gaps before drawing.
The gap between beginners and experienced users is exactly this line: it thinks of the blanks you did not think of. Use it on every prompt, especially when you ask AI to write the prompt for you (run C depends on it).
The safest two-step path for beginners
- Fill the five slots in a chat assistant first: describe the need to a conversational assistant and ask it to write the prompt using the five parts, asking you three questions before finalising;
- Then take that prompt to an image tool: paste it in, lock the ratio, generate twice and pick one.
If you cannot write prompts, let AI write them for you — the success rate doubles. If you cannot write prompts at all, you can also iterate inside the chat assistant directly, describing changes as you go.
Key points from this lesson
| Point | In one line |
|---|---|
| Learn the four terms | Text to image, prompt, reference image, text rendering — no terms, no order |
| Match the six scenarios | Do not learn all six; start with the one you must deliver next week |
| Five parts are the skeleton | Subject ＋ scene ＋ style ＋ composition ＋ text — miss one and it drifts |
| Ratio is a placement decision | 3:4 / 16:9 / 9:16 / 1:1 — get it wrong and the image cannot be used |
| Text is mandatory | Without it you get a beautiful blank illustration |
| Ask three questions | End with "ask me three questions" to turn an instruction into a conversation |
Full Visual Sets with AI, No Design Skills Needed
One five-part formula, 3 hands-on runs and 8 templates — turn posters, article covers and slide visuals into publish-ready images in 20 minutes.
Hands-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.