Why do I need to ask for transparency twice?

The parameter controls the file. The sentence controls the design. Without the sentence you get a correctly transparent file containing a mark that was composed against an implied background.

I asked for transparent and got a checkerboard. What happened?

The model painted the checkerboard as a picture of transparency. It is opaque. Add "no checkerboard" to the prohibition list — the official prompt includes it for exactly this reason.

Which formats keep the alpha channel?

PNG and WebP. JPEG does not, and it will flatten silently. For PNG, leave output compression off.

Is generating four at once cheaper than four runs?

On credits it is the same — the count multiplies linearly. What you save is time and the drift between runs, which matters when you are comparing options rather than iterating.

GPT Image 2.5 prompt — a transparent-PNG logo, four variants in one run

The prompt asks for transparency and the API asks for it too. Both, or you get a drawn checkerboard.

Settings the Get four logo options on a genuinely transparent background prompt ran at
Mode	Model	Quality	Size	Ratio	Credits	Background
Text to Image	Think	Standard	1K	2:3	24 cr	PNG

Published by OpenAI as an example for this model — not generated on this site. Model: gpt-image-2.5-sunburst. Source: OpenAI The published example ran at 1024x1536, a legal size we do not currently offer; the frame here is the same shape at our nearest size.

A mark for a fictional bakery called Field & Flour, delivered as four variants with real alpha. The thing to steal here is not the bakery — it is the pairing. Transparency has to be requested in the prompt *and* set as a parameter, and the prompt has to explicitly forbid the model from painting a checkerboard, because a checkerboard is what transparency looks like in every screenshot it has ever seen.

In full

The prompt, nothing cut

Create an original, non-infringing logo for a company called Field & Flour, a local bakery.
The logo should feel warm, simple, and timeless. Use clean, vector-like shapes, a strong silhouette, and balanced negative space.
Favor simplicity over detail so it reads clearly at small and large sizes. Flat design, minimal strokes, no gradients unless essential.
Fully transparent background. Deliver a single centered logo with generous padding, clean alpha edges, and no solid backdrop, scenery, checkerboard, or watermark.

520 characters

Run it here

This exact prompt, at these exact settings

The mode and all five settings are locked to this template, so there is nothing to line up by hand. Press Run this prompt above and the panel opens filled in — prompt, mode, quality, size and frame — and your first image needs no account. If your plan sits below what this ran at, the line under the panel says so.

This example was published at Think · Standard · 1K. A free run is Fast · Standard · 1K, so expect less fine detail and softer small text. The composition is the same; the finish is not.

3 levers

What is safe to change

Most libraries publish this list and stop, which is why so many copied prompts come back worse than the original: the swappable nouns are the safe half. The three sections under it are the other half — the clauses that are doing the work, the ones that break it if you touch them, and why it is written in this order.

The legibility constraint
"Favor simplicity over detail so it reads clearly at small and large sizes" is what separates a logo from an illustration. It is also the line most people leave out, which is why most AI logo output falls apart at favicon size.
The variant count
Four in one run. A logo decision is a comparison, not a generation — you want options side by side. Raising the count is the cheapest way to make the run useful, and it multiplies the credit cost exactly linearly, nothing hidden.
The brand adjectives
"Warm, simple, and timeless" is the whole brief. Three adjectives is about the right budget: enough to point somewhere, few enough that they do not contradict each other. Replace them and keep everything else.

Three ways to break it

Each one is a real failure with a reason attached. A rule without a reason is not usable.

01Setting the parameter but not saying it in the prompt
Transparency needs both. With only the parameter, the model frequently composes the mark against an implied backdrop and you get alpha around a picture that was designed to have a background. The prompt sentence is what makes it design an isolated mark.
02Exporting as JPEG
JPEG has no alpha channel and does not tell you it dropped one — it silently flattens to white. Use PNG or WebP, and if you use PNG, do not set output compression.
03Dropping "no checkerboard"
This is the single most GPT-Image-specific gotcha there is. The model has seen transparency represented as a grey-and-white checkerboard in millions of design screenshots, so asked for a transparent background it will sometimes *paint one*. The result is an opaque image of a checkerboard. Name it as forbidden.

Why it is written this way

Inside the Get four logo options on a genuinely transparent background prompt

Two things make this template worth its row, and neither is the bakery.

The first is that transparency on this model is a two-sided request. There is an API parameter for it, and there is a sentence for it, and the parameter alone is not enough. The reason is compositional rather than technical: a model that is not told the mark will be isolated tends to design something that assumes a surface — a badge with an implied paper edge, a lockup balanced against a background colour. You then get a correctly transparent PNG of an image that was never meant to float. The sentence "Deliver a single centered logo with generous padding, clean alpha edges" is what changes the design brief, not just the file format.

The second is the checkerboard clause, and it deserves to be famous. Every designer knows the grey-and-white checkerboard as the *convention* for showing transparency. So does the model, because it has seen it behind cut-out assets in an enormous number of screenshots and tutorials. Ask for a transparent background without ruling it out and you will periodically receive a fully opaque image with a checkerboard painted into it — a picture of transparency rather than transparency. OpenAI names this trap in its own guidance: a drawn checkerboard is not transparency. Four words in the prompt prevent it.

Beyond those two, the prompt is conventional and that is the point. "Clean, vector-like shapes, a strong silhouette, and balanced negative space" is ordinary design language, and it works because it is ordinary — these are the terms attached to good marks in the material the model learned from. The one line worth copying verbatim into any logo prompt is the legibility constraint, because scale-independence is the actual functional requirement of a logo and it is the requirement that generative output most reliably fails.

A practical note on checking your result: open the PNG and look at the alpha channel, not the preview. A viewer with a white background will show a white-backed image and a transparent one identically.

4 questions

Get four logo options on a genuinely transparent background — common questions

Why do I need to ask for transparency twice?: The parameter controls the file. The sentence controls the design. Without the sentence you get a correctly transparent file containing a mark that was composed against an implied background.
I asked for transparent and got a checkerboard. What happened?: The model painted the checkerboard as a picture of transparency. It is opaque. Add "no checkerboard" to the prohibition list — the official prompt includes it for exactly this reason.
Which formats keep the alpha channel?: PNG and WebP. JPEG does not, and it will flatten silently. For PNG, leave output compression off.
Is generating four at once cheaper than four runs?: On credits it is the same — the count multiplies linearly. What you save is time and the drift between runs, which matters when you are comparing options rather than iterating.