34Views 0Comments
Ask an AI for a planner cover with the word Planner across the top, and for a long time what came back said Plnaner. Or Plannner. Or something that was not a word in any language.
That was real, and it lasted long enough to become permanent advice. Never put words in an AI image. Generate the picture blank and type the title in Canva. You will still find that advice everywhere, and it is now about two years behind the tools.
What actually changed
OpenAI has released new image models for ChatGPT three times since that reputation formed: in December 2025, in April 2026, and again on 8 September 2026.
The December one matters most here. Its announcement carried a section headed Text rendering, and OpenAI wrote that the model “takes another step ahead in text rendering, capable of handling denser and smaller text.”
Note what that sentence does not say. It does not promise perfection. It says the model got better at something it used to be bad at, which is a fair description of where things stand.
Check it yourself: OpenAI’s announcement of the December 2025 image model. Scroll down past the first set of sample pictures and look for the heading Text rendering. The two 2026 releases are listed in the ChatGPT release notes, in the dated entries for 21 April 2026 and 8 September 2026.
Where the old advice came from
The complaint threads people still link to are from 2025. The one that gets quoted most was posted on 16 June 2025 by someone who could not get a single clean word out of four different tools.
Another user suggested asking for the text one line at a time instead of in a block. It worked, and the original poster said so. Good advice for June 2025, written about models that have since been replaced three times.
Check it yourself: the original thread on OpenAI’s community forum. The date sits directly under the username at the top of the first post.
How to ask for text now
Write the words exactly as you want them to appear, put them in quote marks, and say where each piece sits. Quote marks matter, because they tell the model which part of your sentence is an instruction and which part is the actual text.
A planner cover in soft sage green, with empty space across the top third.
Across the top, large and bold: “My 2027 Daily Planner”
Underneath it, smaller: “One Page A Day”
Short lines still work better than long ones. That is less about spelling now and more about design, because a fifteen word sentence stretched across a book cover looks wrong however perfectly it is spelled.
The habit worth keeping
Read the words on the finished image before you put the file up for sale. Out loud, letter by letter, not a skim.
Your eyes quietly repair small spelling errors while you read, which is how a typo survives four people looking at a cover and turns up on a printed book. This was true before AI existed and it has nothing to do with which tool made the picture.
When to still type the words yourself
There is a good reason to keep using Canva, and spelling is no longer it.
Words baked into a picture cannot be edited afterwards. If you want twelve covers sharing one design with twelve different titles, or you need to change a subtitle after launch, or you sell the same planner in English and French, those words have to live on a layer you can still reach.
For a single cover you will make once, let the AI write the words. For a design you plan to reuse, keep the picture and the text apart. That is a decision about your workflow, not about whether the machine can spell.
