TutorialsAssignment: From Reproducing Three Images to Building Your Own Workflow
Three progressive parts in one assignment — reproduce the three in-class images with the course brief, run one real image you need this week through the full process, then turn it into three role-specific templates. All three required.
Goal
Connect the three things you learned and hand in a workflow you can use tomorrow:
- Reproduce: using the course brief, re-run the three in-class images (poster / article cover / slide visual) yourself;
- Apply: take one image you genuinely need this week and run it through the full process — write the prompt, generate, fix the text;
- Advance: turn the recurring image needs of your role into 3 dedicated templates and a repeatable workflow.
What earns marks is not how pretty the images are, but whether you can explain why the prompt was written that way and what you changed.
Requirements
Hand in three things: three reproduced images with their prompts, one complete record of a real image, and three role-specific templates. All three required — missing one means incomplete.
- Use a different subject for part two: the reading event is off limits there. Use something you actually need — an event poster, an article cover, a Moments image or a deck visual.
- Submit the prompts themselves: attach the prompt you actually used for each image. Do not submit finished images alone — a wrong word on the image is instantly explained by the prompt.
- Write down your edits: what you changed in the AI output and how you phrased it, item by item.
- Format: one document or paste directly into the submission area; images and prompts must correspond one to one.
- Deadline: parts one and two within 48 hours of the lesson; the template set in part three within 7 days (it needs two or three real uses).
Run the five red lines first: read every date, time and venue on the image; anonymise real projects (no client names, internal data or real faces); the AI output is a draft, so review it yourself before publishing; no real celebrity faces and no third-party logos; label AI-generated images when publishing publicly.
Part one: reproduce (three images from the course brief)
Run all three using the course brief, then submit the images together with their prompts.
| Item | My situation |
|---|---|
| The image tool I used | |
| A poster: was the ratio 3:4 | |
| A poster: do the date, time and venue on the image match the brief | |
| A poster: how many re-rolls did I do, and why did I pick the last one | |
| B cover: is the headline still readable at thumbnail size | |
| B cover: is the text area free of overlapping artwork | |
| C slide visual: what metaphor did the prompt AI returned use | |
| C slide visual: did any text appear on the image (this one should have none) | |
| Which parts of the three images I edited, and with what phrasing | |
| The question I asked back, plus the AI answer |
Rows 3 and 8 are the key observations for this reproduction.
- Row 3: AI getting dates, times or venues wrong is a high-frequency event, and it never warns you. Zoom in and check character by character. If something is wrong, fix only that item with template 7's phrasing ("keep the composition and palette, change only the small detail text to …") instead of rewriting the whole prompt.
- Row 8: the constraint for the deck visual says no text. If text or numbers still appear, the constraint sits too early — move it to the final line, run again, and submit both results.
Self-check
- The three ratios are 3:4, 16:9 and 16:9 — none mixed up?
- Every image has its prompt attached, not just the picture?
- Every word on the images was read, and the dates and times are correct?
- The cover is still readable at a very small size?
- Edits used "keep … change only … leave the rest unchanged", not a rewritten prompt?
Part two: apply (one real image you need this week)
Pick an image you will actually use. Running an event — use the poster. Writing a post — use the cover. Preparing a deck — use the slide visual. It must be for real use, not practice.
| Item | My situation |
|---|---|
| What this image is for (purpose ＋ placement) | |
| Who sees it (group / client / leadership) | |
| My seven-row brief sheet (copy it) | |
| The joined prompt (copy it) | |
| How many generations, and which one I picked | |
| My first edit ＋ the phrasing I used | |
| My second edit ＋ the phrasing I used | |
| Total time spent | |
| How long the same image used to take | |
| Did it go out (where / what is still missing) |
Do not write "tweaked it a bit" in the edit rows. Be specific: for example "changed the headline from 读书会 to 全民阅读月 using template 7, and touched nothing else" or "swapped the background from day to night, nothing else moved".
Those edit points are exactly what you should add to the constraint slot next time.
Part three: advance (three role-specific templates)
Condense the images your role produces repeatedly into three templates. Do not copy three of the eight — rewrite them in the language of your own job.
| Item | My situation |
|---|---|
| The three most frequent image needs in my role (one line each) | |
| Template 1: use case ＋ full prompt | |
| Template 2: use case ＋ full prompt | |
| Template 3: use case ＋ full prompt | |
| Which parts are the fixed skeleton across all three | |
| Which parts are the variables each time | |
| Where I stored the templates (tool custom instructions / document / notes) | |
| My workflow in one line (who writes, who generates, who signs off) |
Do not build an elaborate system on day one. At minimum do two things: store the three templates in the AI tool you use (custom instructions or an agent if it has them), and create one folder per image type for finished files.
Using them three times beats building something perfect and forgetting it.
Grading
| Level | Standard |
|---|---|
| ⭐⭐⭐ Excellent | All three parts; ratios and text on all three reproductions correct; the real image run through the full seven-row brief → prompt → generate → edit chain with specific edit notes; the three templates genuinely fit the role and clearly separate fixed skeleton from variables |
| ⭐⭐ Good | All three parts; three reproductions succeed and are directly usable; the real image has a complete prompt and edit record; three templates take shape |
| ⭐ Entry | All three parts; three usable images reproduced from the brief with their prompts attached |
The two usual reasons for missing ⭐⭐⭐: the small detail text was never checked (a wrong date or venue slipped through), and the templates are just the eight from class copied over — not rewritten in your own job's language, so they will not be used again.
Submission
- Upload a screenshot of each finished piece to the site as your submission.
- Pick the one you are happiest with and post it in the course group so everyone can see your workflow.
One last reminder: read every word on the image before it goes out. AI gets dates and venues wrong with more confidence than you expect — and the image carries your name.