How to Make a Poster with GPT Image 2.5: A Cinnamon Roll Brief

I've been thinking about a bakery brand called melo.

The lowercase name matters to me. I like a white background, blue lettering, and a little of the looseness of handwriting. Simple and a bit cute, without looking like it's made for children. As for the product, I have a soft spot for a classic cinnamon roll. No frosting.

That gives us a fairly specific GPT Image 2.5 poster exercise: introduce a cinnamon roll in a portrait layout, make the product easy to recognize, and give people a chance to remember the name melo.

This is a concept brief. There is no real opening date, shop address, or offer behind it. The illustration is AI-generated to show the design direction; this prompt has not yet been tested in PixOnDeck with GPT Image 2.5. The image is not evidence of that model's output.

Decide what this poster needs to say

I'd keep the content short: melo is the brand, the cinnamon roll is the subject, and the poster needs a product name.

For now, I'd leave out the brand story, ingredients, price, and ordering information. Another image might need those details. This one has a smaller job: make someone interested in the roll.

I also wouldn't translate “New York style” into yellow taxis, graffiti, or a pile of English stickers. What interests me is the feeling of a small bakery with a clean visual identity and a personality of its own.

So the white and blue stay. The roll keeps its natural baked color. There isn't much reason to add anything else.

Choose the model, then give it a brief

In the PixOnDeck generator, select GPT Image 2.5 and a 2:3 portrait aspect ratio, then enter the prompt below. Set the ratio in the controls as well as mentioning it in the prompt.

Create a 2:3 portrait concept poster for a bakery brand called melo.

Purpose:
Introduce a classic cinnamon roll. The viewer should notice the product
first, read the product name next, and remember the brand.

Subject:
One classic cinnamon roll, with clearly visible spiral layers
and a baked surface.
Keep its natural golden-brown color. Do not add cream, frosting,
fruit, or any other toppings.
The roll is the only food in the image. No coffee or other pastries.

Composition:
Place the cinnamon roll in the lower-middle part of the poster,
occupying the main visual area.
Arrange the headline above it and leave generous space around both.
Do not add a table setting, people, utensils, or a detailed scene.

Color:
A clean white background.
Use a saturated blue for the brand name and headline,
with a visual direction close to Klein blue.
Keep the food naturally colored.

Text:
Include only these two pieces of text, spelled exactly:
“melo”
“cinnamon rolls”

“melo” must be entirely lowercase and smaller than the product headline.
Use “cinnamon rolls” as the main headline, clearly legible.
Do not add dates, prices, addresses, slogans, or any other text.

Overall direction:
Simple, with a relaxed handwritten feel.
Avoid a children's illustration style, ornamental vintage borders,
promotional badges, and unnecessary decoration.

It's a fairly long prompt, but each section makes a decision that can be checked in the result.

“Make it well designed” is hard to assess. “One cinnamon roll,” “lowercase brand name,” and “no frosting” give me something concrete to look for.

Without a product photograph, this is only suitable for exploring a concept. To advertise an actual product, I'd start with a real photograph and check whether the result has changed the product.

What I'd check in the first image

First, the food itself. Are the spiral layers clear? Has anything been added? Does the surface look baked, or shiny enough to be plastic? A generous layer of frosting might look delicious, but it would rule the image out for this brief.

Then I'd view it at roughly the size someone would see on a phone. Can I still read the product name? Is the roll still the subject? Has melo ended up so small, or so crowded by decoration, that it barely registers?

Only then would I look more closely at the letter spacing, edges, shadows, and the particular shade of blue. OpenAI’s documentation also notes limits in text clarity and precise placement. Text that seems fine at a glance still needs checking for spelling, letter shapes, and placement.

AI-generated concept illustration showing the intended relationship between color, product, and text. It is not a verified GPT Image 2.5 result from PixOnDeck or a photograph of an actual product.

If the product works, make the next request smaller

Suppose the roll is right but the headline is too big. I'd download that result, upload it again to the generator, and give it a focused edit request:

Edit the uploaded image.

Only reduce the size of the “cinnamon rolls” headline,
leaving more space between the headline and the roll.
Keep the headline's wording, color, and type style.

Preserve the cinnamon roll's shape, position, surface texture,
and lighting.
Keep the spelling, lowercase letters, and position of “melo”.
Do not add any new elements.

“Preserve” is a request, so I'd still check the whole result afterward. For a closer look at how to describe the relationship between a headline, an illustration, and empty space, see this poster prompt exercise.

If the food and composition work but the type won't sit where I want it, I'd finish the lettering in a design tool. With an existing logo, I'd use the actual logo file instead of asking a model to redraw something similar.

For this exercise, the finish line is fairly clear: the product follows the brief, the text is correct, the layout reads at a small size, and nothing unnecessary has crept in. After that, I can decide whether another style adjustment is worth making.