Experimenting With AI: Designing a Seed-Packet-Style Crop Timing Diagram Using ChatGPT

/ ChatGPT Prompts / By

If you are gardening in cool or cold climates, timing matters more than almost anything else. A few weeks too early or too late can mean poor germination, stunted growth, or missed harvests. Seed packets try to help, but they rarely show the whole picture. In this post, I’ll show you how I designed a clear, seed-packet-style crop timing diagram that summarizes seeding, planting, growth, harvest, watering, and fertilizing all at a glance.

I an idly thinking I‘d like some nice diagrams for my new gardening blog, so I open ChatGPT and type:

I want to create a prompt that when run, creates a diagram that summarizes the main timing factors when seeding, planting, growing and harvesting a CROP. I want it to apply to the northern hemisphere in places with a cool climate. Ask me questions to help refine what I want. Ask your questions one at a time, awaiting my answer before proceeding.

ChatGPT gets right on it, starting with a little encouragement before asking for clarifications:

Great idea! — this kind of diagram can be very powerful for gardeners 👍
Before I write the final reusable prompt, I’d like to clarify a few things so it does exactly what you want. I’ll ask one question at a time, and wait for your answer before continuing.

Question 1

What should the final output of the prompt produce?

Please choose one (or say “other”):

  1. A text-based diagram (using characters like ▬ █ ░) that displays directly in chat
  2. Mermaid diagram code (for rendering elsewhere)
  3. A prompt for an image generator that creates a visual timeline graphic
  4. Something else (please describe briefly)

What’s your choice?

3

Got it 👍 — we’re designing a prompt for an image generator that will create the diagram.

Question 2

Do you want the diagram style to be…

And so on, for a whole set of detailed questions that refine what my intention is and what my needs are. I like to use this work method — “build me a prompt but ask me questions first” — when I know I am going to need a robust prompt that I will use multiple times.

Some of ChatGPT’s questions are quite tricky. For example, I am asking for barcharts that cover some basic plant needs, and ChatGPT asks:

Question 7

How should watering and feeding be represented in their rows?

Please choose the style you prefer:

A. Intensity bars
– Thicker / darker bars during high-need periods
– Lighter / thinner bars during low-need periods

B. Simple on/off bars
– Bar present = needs attention
– No bar = minimal or no need

C. Icons + bars
– Bars for timing, small watering-can / fertilizer icons to show emphasis

Which option fits your vision best?

What has previously been just a vague notion in my head — showing how and when to water — suddenly becomes very concrete. What really do I want? None of the suggestions being offered really fit my vision. After a lot of thought I write:

Assume 1 watering per week. Use narrow bars, their height indicating watering amount (1cm, 2cms...). Keep the bar chart compact. For water, give scale on left of chart, showing amount of water to give per week: 1cm, 2cm, etc. Bars do not need to be that actual height, keep them small, for example 1mm, 2mm, etc. Fertilizing should have a similar chart, but no scale needed. Do you understand? Ask me more questions if you need clarification.

I think this highlights a major benefit of the “build me a prompt but first ask me questions first”. I hadn’t really thought through how I wanted the watering and feeding information presented at all, but now I realized it was a key thing I need the diagram to show clearly.

Towards the end of the question, ChatGPT gives me a summary of what it thought I needed:

My understanding so far

  • The output will be an image-generator prompt.
  • The image is a seed-packet-style timing diagram for a single crop.
  • Climate: Northern Hemisphere, cool-temperate.
  • X-axis: Months (Jan–Dec), shown as three-letter abbreviations.
  • Layout: ….

Finally, ChatGPT is ready and created the prompt. I won’t put the whole thing here because it is rather long and detailed. You can find the latest version at the foot of this post. (And if you want to recreate it yourself you can always follow the earlier steps and experience ChatGPT’s questioning session for yourself.)

I was eager to run the prompt and see what the output was.

run the prompt for the crop beetroot
First draft of the crop diagram for beetroot generated by ChatGPT

 Wow! That was a lot better than I expected!

But of course I wasn’t completely satisfied, so after a lot more back and forth, requesting minor tweeks and major changes, I eventually persuaded ChatGPT to rewrite the prompt so that the output more closely matched my likes.

This involved such marvelously convoluted instructions as:

Change the prompt so that only boreal climate (blue bars) and cool-temperate climate (green bars) are shown. Calculate the position and range of the cool-temperate bars first, paint them, then calculate the position and range of the boreal bars and lay them on top of the cool-temparate bars, so the resulting bar has a blue center and green ends.

Running the nth version of the prompt resulted in what I felt was very close to the final goal:

Updated beetroot diagram showing both boreal and cool-temperate bars. But with some mysterious extra bar

So much better! Providing the data is accurate, this is going to be really useful — it is something I can show for each of the 70+ crops I have already covered in my blog. There is just one little thing, nothing serious, not really a problem at all…

The first six rows have bars and a horizontal line through them. What is the line for?

ChatGPT gets excited.

Good catch — that horizontal line is not something you asked for, and it isn’t conveying any extra data.

What the line actually is

That line is simply a row separator / baseline that the image model added automatically to structure the chart. In other words:

  • It is not a data signal
  • It does not represent duration, threshold, or intensity
  • It is not tied to climate, timing, or growth stage

Think of it as the equivalent of ruled paper lines or a table grid line.

Why it appeared

Image generators often:

  • Add horizontal rules to keep rows visually aligned
  • Treat each row like a “track” with a baseline
  • Do this especially in infographic or chart-style prompts, even when not requested

Because your prompt specifies:

  • “multiple horizontal rows”
  • “compact, instructional chart”
  • “seed-packet style”

…the model inferred light row dividers to improve readability.

ChatGPT continued on with this justification for a lot longer, eventually offering me several solutions which would solve the problem once and for all. Needless to say, it didn’t. So far, none of the modifications I have made to the prompt eliminate those annoying artifacts. I suspect ChatGPT doesn’t know where they come from itself, and it’s just casually bullshitting answers at random. I think I will get there in the end with a clean end result, but that’s for another day.

Here is the current version of the prompt. Try it for you favourite vegetable, fruit or herb.

SHORT IMAGE GENERATION PROMPT (DAILY USE — TWO-CLIMATE OVERLAY)

Create a clean, seed-packet-style timing diagram for [CROP NAME] for Northern Hemisphere cold climates, comparing cool-temperate and boreal regions only.

Style

  • Flat, technical infographic
  • Minimal, compact, instructional
  • Black axes and text on light background
  • No decoration, no paragraphs
  • Designed to look like a professional seed packet chart
  • Precise, calm, and readable at a glance

Layout

  • Landscape orientation
  • Horizontal chart
  • Months along top:
    Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
  • Multiple horizontal rows
  • Bars are narrow and compact
  • Bar length = timing

Rows (top to bottom — exact order)

  1. Indoor seeding
  2. Transplanting
  3. Direct sowing outdoors
  4. Active growth
  5. Flowering / fruiting
  6. Harvest
  7. Watering
  8. Fertilizing

Climate overlay logic (critical)

  • Only two climates are shown:
    • Cool-temperate = green
    • Boreal = blue
  • For each row:
    1. First calculate the full timing range for the cool-temperate climate
      • Paint this bar first, in green
    2. Then calculate the timing range for the boreal climate
      • Boreal timing is generally later and shorter
      • Paint this bar on top of the cool-temperate bar
  • The resulting visual effect:
    • Blue center section (where boreal overlaps cool-temperate)
    • Green extensions at one or both ends (cool-temperate only)
  • Bars must be perfectly aligned horizontally so the overlap is obvious
  • No separate climate rows and no separate legend needed; the color overlay explains the relationship

Climate assumptions

  • Cool-temperate: earlier start, longer season
  • Boreal: later start, shorter season
  • Timing must be biologically realistic for the crop

Watering row

  • Assume 1 watering per week
  • Very narrow bars
  • Bar height = water per week
  • Include a left-side vertical scale labeled:
    • 1 cm
    • 2 cm
    • 3 cm (if needed)
  • Bars are visually scaled down for compactness
    (e.g. 1 mm bar height = 1 cm water per week)
  • Apply the same green-first, blue-overlay logic as above

Fertilizing row

  • Narrow bars
  • Bar height = feeding intensity
  • No numeric scale
  • Apply the same green-first, blue-overlay logic

Text

  • Only include:
    • Crop name title
    • Row labels
    • Month names
  • No legend
  • No icons