For a one-off image, make it with GPT Image 2.5 in ChatGPT.

Why GPT Image 2.5

The reasons it wins for this job.

On Artificial Analysis's blind text-to-image arena, GPT Image 2.5 Flare ranks first (Elo 1186) and Sunburst second. On the editing arena, Sunburst ranks first (1176).
Editing across turns is how a one-off image gets finished: "same shot, warmer light, move the logo left" without starting over.
OpenAI positions Sunburst for "production-ready campaign creative or polished product imagery", which covers most marketing one-offs.
It renders legible text in images well enough for headlines. For dense typography, see Ideogram below.

How to use it

4 steps to a first result.

1Describe the shot, not the vibeSubject, setting, lighting, camera angle, aspect ratio. "Modern and clean" tells it nothing.
2Upload a referenceA product photo or brand example keeps colours and shapes right.
3Edit instead of regeneratingAsk for specific changes to the image you have. Regenerating loses what was already right.
4Export at the size you needAsk for the aspect ratio of the placement: 1:1 for feeds, 1.91:1 for link cards.

Prompt for ChatGPT

Start from this.

Edit the parts in capitals, then run it.

Create a SUBJECT image for PLACEMENT (aspect ratio RATIO).

Scene: WHAT IS IN THE FRAME, WHERE.
Lighting: e.g. soft morning window light.
Camera: e.g. eye level, 50mm, shallow depth of field.
Style: photographic, not illustrated. Muted colours with one accent of BRAND COLOUR.
Text in the image: exactly "HEADLINE" in a clean sans-serif, top left. No other text.

Use the attached image as the reference for the product's shape and colour.

Nothing is sent anywhere. It copies to your clipboard.

Alternatives that also work

If GPT Image 2.5 is not an option.

Nano Banana 2 (Gemini 3.1 Flash Image)Official page →

Google · closed · cost: low

Pick it when you work in the Gemini app, or want many quick variations. It ranks sixth on the same arena.

Ideogram 4.0Hugging Face →

Ideogram · open weights · cost: low · 64K HF downloads/mo

Pick it when the image is mostly text and layout, such as a poster or packaging. Ideogram is built for multilingual typography and layout control.

Qwen-Image 2512Hugging Face →

Alibaba Qwen · open weights · cost: free weights; you pay for the hardware · 53K HF downloads/mo

Pick it when you want to run it yourself with no usage limits. Apache 2.0.

Watch out for

Real people, real logos and trademarked characters are restricted on most platforms and legally risky everywhere. Use your own assets as references.
Midjourney has no public API and its current version is not ranked on the arena, so it is absent from this comparison, not ruled out.

Doing this again and again? For images at volume, generate them with Nano Banana 2. Product shots, ad variants and article images at scale need good quality at a predictable cost per image. Nano Banana 2 ranks sixth on the text-to-image arena at roughly a third of the leader's price per image, from a vendor with stable API terms.

Sources

Checked 17 September 2026. Models change monthly; we re-check this page when they do.

Picking the model is the easy part.

Wiring it into a workflow that runs every week, with evals, fallbacks and a cost you can predict, is the work. Fifteen minutes, no deck.

Book fifteen minutes →