Six lines went into the checking. Three came out.
That ratio is the whole job. A birthday poster of this kind is a short list of things that were true on one particular day, set in a retro frame, and the part that takes the time is not the picture. It is deciding which of the lines you want to print you can actually stand behind. This piece works one date all the way through, then generates the poster twice to see what the text does.
The finished poster. Each line on it was checked against a source before it was written into the prompt.
Six lines in, three lines out
The date is 12 March 1994, picked arbitrarily and attached to nobody. The candidate lines were the obvious ones for this format: a chart-topping song, a news event, a film, a price, a second chart from another country.
Each line was looked up before it went anywhere near a prompt.
Three survived:
Three did not. The US box office came back with two different films named as number one for that weekend by two sources. The UK singles chart came back with a disagreement about which week the date belongs to. The average US petrol price was not settled by the sources I looked at in the time I gave it.
What makes a line checkable
The three failures have the same shape, and it is worth naming because it decides how you build the list. A chart has a week. A birthday has a day. Those two things do not line up, so any line that starts "the number one X" carries a question about which week you mean before it carries a fact — and different sources answer that question differently.
The lines that survived are the ones pinned to the day itself, or to a rate that was in force across it. An ordination that happened on a dated morning is checkable. A postage rate that ran for four years covers any day inside it. A chart position needs you to decide, and print, which convention you used.
So the practical rule for building the list: go looking for things dated to the day, and for things that were simply true for a stretch of time containing it. Treat every chart line as costing twice as much work as it looks, and be ready to drop it.
Write the checked lines into the prompt, not the topic
Once the three lines exist, the prompt is short, and the important thing about it is that it contains the finished text rather than a description of it. Nothing in the request asks the generator to know anything about 1994.
A retro hand-drawn commemorative poster on aged cream paper with a soft grain and worn edges, muted ink colors, letterpress feel, no photographs. At the top, large hand-lettered heading: "12 MARCH 1994". In the middle, a simple hand-drawn illustration of a small terraced street under a rain cloud. Along the bottom, three short lines of hand-lettered text, each on its own line, spelled exactly as written: "NO. 1 SONG: THE SIGN - ACE OF BASE" / "FIRST 32 WOMEN PRIESTS ORDAINED IN ENGLAND" / "A FIRST-CLASS US STAMP COST 29 CENTS". 16:9 landscape.
It spelled the birthday poster correctly four times out of four
The same prompt sent four times through the image mode of the video composer, one credit each.
That prompt carries about a hundred and ten characters of specified text. Sent four times at one credit each, all four came back with every string spelled correctly — the date, the song, the ordination line, the stamp line. What moved between runs was the line breaking and whether the dash in "THE SIGN - ACE OF BASE" came out as a hyphen or an en dash.
This is worth saying plainly because the usual advice for this kind of poster is to leave text out of the prompt and add it afterwards in an editor. On this model, on this date, with text supplied verbatim in quotation marks, that was not necessary.
What it would not do is put the lines where I asked
The brief said "three short lines of hand-lettered text, each on its own line, along the bottom". In four runs out of four it set them as three columns across the foot of the poster instead. The words were right in each run and the arrangement was wrong in each run.
Sixteen credits bought the layout, not the spelling
The workspace has a second generator for this kind of work, alongside the AI video generator used elsewhere in this series. Design Studio sits above Video Studio in the left rail, takes the same kind of prompt, and prices a generation at 16 credits rather than 1. The same poster text went in unchanged.
Left, the one-credit version. Right, the sixteen-credit version. Same words in both.
It put the three lines where the brief asked: stacked, each on its own line, along the bottom. The spelling was correct, as it had been at one credit. So the fifteen-credit difference did not buy correct words. It bought the arrangement of them, plus a canvas you can keep working on — the result opens in an editor with Remove BG, Upscale, Eraser, Expand, Recolor and an Edit text control, and a side panel offering to adjust the text placement again.
Which of the two to use is a question about how much the layout matters to you, not about whether the generator can write. If three columns of small text along the foot of a poster is fine, that poster costs 1. If you need the stacked list, the same poster costs 17 by the time you have both.
What the poster cost
The cost lines as shown before each request, read on 15 September 2026.
Twenty-one credits, and the balance in the header moved from 23,925 to 23,904, which matches. The stills were made through the AI image generator flow. These figures are for these configurations on this account and this date; other models and modes are charged differently, and prices change.
Set that against the other half of the job. The generating took twenty-one credits and a few minutes. The checking took six lookups to produce three printable lines, and that is the part that decides whether the poster is worth giving to anybody.
Price the job by the checking, not by the prompt
The one-credit result, for comparison: same words, different arrangement.
If you make one of these for somebody, the estimate that matters is not the credit count. It is how many lines you are promising and how long each one takes to stand up. Three checkable lines took six lookups here, and the three that failed were not obscure — they were the chart positions, which are the first thing anyone reaches for.
So quote the work in lines, not in generations. Pick the date, spend the hour finding things dated to the day, throw out anything two sources disagree about, and only then open the composer. The poster is the cheap end of this, whichever generator you use; the list is the thing you are actually making.
The date used here is real, and the person is not — nobody is named on the poster and no birthday of any actual individual was looked up to produce it. Keep it that way round in your own version: build the list from public, dated sources, and do not assemble a page of personal details about someone who has not asked you to.
Print the year on each line that carries one, so a reader can check it. If a line came from a single source that you could not corroborate, either leave it off or say where it came from — a poster is a thing people keep, and an invented fact on it lasts as long as the frame does. The generator was not asked to supply any of the history here, and asking it to would remove the only step in this process that makes the result worth anything.
Also not covered: printing and paper sizes, animating the finished poster into a greeting clip, multi-page sets, and dates outside the range where English-language sources are easy to search.
Written 15 September 2026. The poster illustration and lettering were generated in CapCut; the three facts on it were taken from the sources linked above and are quoted with their year. The four one-credit runs are 2560 by 1440; the Design Studio result is 2752 by 1536. "Spelled correctly" means every character of each quoted string in the prompt appears in the rendered poster, checked by reading the generated images; line breaks and dash characters were not required to match. Interface labels, the credit chip and the cost lines reflect a single session on that date and the configurations named above, and may change. The one candidate line about petrol prices was dropped for lack of a settled source rather than because no source exists.