Depikt
Back to blog
Images 2.56 min

Posters, Infographics, and Slides: Structured Visuals with ChatGPT Images 2.5

Images 2.5 handles defined layouts and real-world content better than its predecessor. That only helps if the prompt defines the layout. Here is how to specify text, hierarchy, data, and transparency for the three formats people ask for most.

By Depikt Team

What OpenAI actually claims

Three lines from the launch matter for structured visuals. Images 2.5 "can handle more complex layouts including transparent backgrounds." Images "that include real-world information have more accurate content." And the model is more likely to retain "the requested visual direction, composition, and individual details" as instructions get more specific, which OpenAI ties to "UI concepts that preserve a given hierarchy, or presentation visuals that fit a defined structure." The system card adds that it "improves infographic accuracy and layout."

What OpenAI does not claim: a text-accuracy percentage, or that the model will design a hierarchy for you. The phrase to notice is "a defined structure." The gains go to prompts that define one.

Everything past this point is Depikt's method. The examples are ours.

The rules that apply to all three formats

  1. Quote every string. Text outside quotes is a description and may be paraphrased. Text inside quotes is content.
  2. Give each string a position and a relative size. "Top, largest" and "bottom-left, smallest" are enough; the model handles the type.
  3. Close the door. End with "No other text, labels, logos, or watermarks." Unlabeled space is where invented copy appears.
  4. Name the ratio. 2:3 for print posters, 4:5 or 1:1 for social, 16:9 for slides.
  5. Pick one medium. Risograph, flat vector, photo. A medium carries a style better than six adjectives.
  6. Keep the count low. Three to five text elements, three to five diagram items. More than that and you are designing a document, not an image.

Posters

The GPT Image 2 era guide to posters with text laid out a six-part structure: ratio, headline, image, palette, finish, grounding details. It still holds. What is new is that 2.5 is more likely to keep a stated hierarchy intact, so state it.

Prompt
Print poster, 2:3 portrait. Layout, top to bottom: headline band (top 25%), full-bleed image (middle 60%), footer strip (bottom 15%). Exact text: "NIGHT SWIM" (headline, largest, condensed grotesk, all caps); "Open-air cinema · Fri 26 Sept · 21:00" (footer, left, small); "Free with a lido ticket" (footer, right, small). No other text. Image: a single swimmer floating in a floodlit outdoor pool at night, seen from above, turquoise water, one strong overhead light. Style: screen print, three inks (turquoise, warm white, near-black), visible halftone, paper texture. Avoid gradients, drop shadows, decorative borders, and any second swimmer.

Notes:

  • The layout is given as bands with percentages. This is the "defined structure."
  • The footer has two strings with left and right positions, so the model does not stack them.
  • "Any second swimmer" is a constraint you only think to add after the model has added one. Add it first.

The ChatGPT Poster template (new with 2.5) gives you a frame to fill: information, design elements, style. It is a fine starting point for a one-off. For brand work, exact copy, or a series, write the full prompt so the text and constraints are yours.

Infographics

The 2.5 launch says real-world content is more accurate and the system card says infographic layout improved. Neither says the model knows your numbers. The rule is simple: everything factual goes in the prompt, in quotes, with a source line.

Prompt
Explainer infographic, 4:5 portrait, for a blog post. Layout: title at top; below it a single horizontal row of four equal panels, numbered 1-4 left to right; each panel has one simple line icon above a two-line caption; a footer line at the bottom. Exact text: title "Where a household's water goes" (largest); panel captions, verbatim: 1 "Toilets · 24%"; 2 "Showers · 20%"; 3 "Taps · 19%"; 4 "Washing machines · 17%"; footer "Source: EPA WaterSense, indoor use" (smallest). No other text, numbers, or labels. Style: flat vector, two colors (deep teal, warm gray) on white, 2px consistent line weight, no gradients, no 3D. Do not draw a house, a pie chart, or decorative water droplets.

Notes:

  • Four items with four numbers, all supplied. If you do not have a source line, you are not ready to make the infographic.
  • The visual form is named ("a single horizontal row of four equal panels"). Left to the model, you may get a pie chart and a house.
  • Percentages are in the captions, so the model does not have to draw a proportional chart, which is where numbers go wrong.

The older infographics guide has more on the five-block structure; it was written for GPT Image 2 and the structure is unchanged.

Slides and presentation visuals

OpenAI's own gallery for 2.5 includes a presentation slide explaining solar flares with a four-step diagram. The prompt shape for a slide is an infographic at 16:9 with less on it.

Prompt
Presentation slide, 16:9, white background, for a company all-hands. Layout: slide title top-left; a horizontal four-step process diagram across the middle (four rounded rectangles connected by arrows, left to right); one-line takeaway bottom-left. Exact text: title "How a support ticket moves" (largest); steps, verbatim, one per box: "Received", "Triaged", "Assigned", "Resolved"; takeaway "Median time to resolve: 2 days" (small). No other text. Style: flat, single accent color (deep blue) with dark gray text, 16px-equivalent consistent stroke, plenty of margin. No icons inside the boxes, no gradients, no clip art.

For a deck, generate one slide per prompt and keep the style clause identical across prompts. Images 2.5 is described as better at holding a visual direction across a series; it will not hold one you did not write down.

Transparent backgrounds

Transparency is named in the launch as one of the "more complex layouts" 2.5 handles. Ask for it in words and export as PNG:

Prompt
Isolated illustration of a single paper coffee cup with a plain white sleeve, three-quarter view. Transparent background: no backdrop, no floor, no ground shadow, no vignette. Flat vector, three colors. Nothing else in the frame.

"No ground shadow" is the clause people forget; a shadow on a transparent canvas becomes a gray smear on whatever you place the image over. In the API, the background parameter controls this directly.

Mistakes specific to structured visuals

  • Letting the model pick the data. It will. Supply every number and the source.
  • Six text elements. You get five right and one wrong, and you cannot tell which until you read them all. Cut to four.
  • Describing type instead of naming it. "Elegant font" means nothing. "Condensed grotesk, all caps" or "high-contrast serif, sentence case" does.
  • A chart with real proportions. Proportional bars and pie slices are where numbers drift. Put the values in captions and keep the chart schematic.
  • Forgetting the close. No "no other text" clause. Something will appear in the empty corner.

Check it before you generate

Paste the prompt into the Critique mode. Text and layout is one of the rubric's essential dimensions; an unquoted string or a missing ratio caps the overall score, which is the point. Or start from a rough idea in the Prompt workspace, which asks for the exact text and ratio before it writes anything. The Library has dozens of poster and infographic prompts from the GPT Image 2 era to study; remix any of them for 2.5.

Sources

Build a prompt that follows this guide.

Paste a rough idea. Get back a structured prompt.

Open Prompt

More in Images 2.5

Images 2.5
ChatGPT Images 2.5: What's New and What It Means for Prompting

OpenAI released ChatGPT Images 2.5 on September 8, 2026: better reference fidelity, edits that leave the rest of the image alone, more consistent multi-turn editing, more accurate layouts, and up to 50% lower latency. Here is the launch, sorted by what it means for the prompt you write.

Images 2.5
How to Prompt ChatGPT Images 2.5

Images 2.5 is better at doing exactly what you say and leaving alone what you do not mention. That changes where the effort in a prompt should go. Here is the six-part structure we use, the mistakes that still waste generations, and five rough ideas rewritten into prompts that ship.