Пређи на садржај
Пријави сеПочни бесплатно

How to write prompts for GPT Image 2.5 Flare

GPT Image 2.5 Flare makes pictures from a sentence and edits the pictures you give it. The first two sections put OpenAI’s own prompting guide in our words, and we did not test them. The last section rests on sample pictures made with this model that we looked at ourselves.

Written from the provider’s documentation and from results we checked ourselves. Last checked November 8, 2026.

Make something with GPT Image 2.5 FlareThe GPT Image 2.5 Flare model page

Name the result, then what can be seen

From the provider’s documentation

  • OpenAI’s guide starts with the result: say what the picture is and what it is for, then its framing and any placement rules. For a complicated picture it suggests labelled parts: scene, subject, details and constraints.
  • It asks for visible detail, such as materials, light, colours and the kind of picture. Camera words work as a look, not as exact optics, and it says to ask for “photorealistic” outright when you want that.
  • For people, it says to describe how they are framed, how large they are in the picture, where they look and what they do with objects.

Text, edits and several pictures

From the provider’s documentation

  • For text, OpenAI says to put the exact words in quotes, say where they go and in what lettering, and say whether other text may appear. Check the spelling afterwards, and use a higher quality setting for small or crowded text.
  • For an edit, say what to change and list what must stay: who is in it, the layout, the light, the labels. Change one thing at a time, feed the last result back in, and say again what must stay if it starts to drift.
  • With several input pictures, number each one and say what it is for, such as the subject, the style or the background.
  • OpenAI lists limits too: text placement can still be off, characters can drift from one picture to the next, and exact placement in layout-heavy designs is hard. It does not promise that a prompt gives the result you asked for.

What our sample pictures showed

From results we checked ourselves

  • Words put in quotes were spelled as written in the pictures shown here, in English, Korean and Japanese: the thin white “QUIET MODE” on the earbud ad, the three menu lines on the Korean menu board, the Japanese price card, the label on the soda bottle and the two lines of the fashion poster.
  • When a prompt did not say whether other text was allowed, the model added some. The menu board got a title and a speech bubble, the bento shop got a sign on the wall, and the infographic got a sentence under each step although the prompt asked for short labels. If you do not want extra text, say so, as OpenAI’s guide advises.
  • The chalkboard edit changed the sign’s words and swapped the cup for an iced drink while the plant, the frame and the counter stayed. Small details moved too: the little drawing beside the words changed and the saucer became a coaster.
  • The other two edits kept the scene recognisable and changed what was asked: the bottle’s label and the light in one, the film look and an orange date in the other. These are the examples we looked at, not a promise that an edit keeps everything.

Real results and the prompts behind them

Results made with this model. The prompts are shown exactly as written, in English.

Matte black wireless earbuds case standing open on wet slate, tiny water drops, soft blue rim light, a clean minimal pro
Made with GPT Image 2.5 Flare

Matte black wireless earbuds case standing open on wet slate, tiny water drops, soft blue rim light, a clean minimal product ad with the words "QUIET MODE" in thin white type

Korean dessert café menu board in pastel chalk listing "빙수 9,000원", "크로플 6,500원" and "말차라떼 5,500원", a bowl of mango bing
Made with GPT Image 2.5 Flare

Korean dessert café menu board in pastel chalk listing "빙수 9,000원", "크로플 6,500원" and "말차라떼 5,500원", a bowl of mango bingsu on the counter in front

Japanese bento shop window display with a handwritten price card that reads "鮭弁当 680円", neat rows of bento boxes, warm e
Made with GPT Image 2.5 Flare

Japanese bento shop window display with a handwritten price card that reads "鮭弁当 680円", neat rows of bento boxes, warm evening light

Glass bottle of yuzu soda on a sunlit marble table with sliced yuzu and ice, the label printed with "YUZU FIZZ" in a pla
Made with GPT Image 2.5 Flare

Glass bottle of yuzu soda on a sunlit marble table with sliced yuzu and ice, the label printed with "YUZU FIZZ" in a playful rounded font and "유자 스파클링" underneath, crisp shadows

Minimal fashion poster with a model in a sculptural white puffer on a pale grey set, the headline "WINTER FORM" in thin
Made with GPT Image 2.5 Flare

Minimal fashion poster with a model in a sculptural white puffer on a pale grey set, the headline "WINTER FORM" in thin wide letters at the top and "COLLECTION 26" small at the bottom

Flat vector infographic poster titled "How Coffee Is Made" with five numbered steps from cherry to cup, simple icons and
Made with GPT Image 2.5 Flare

Flat vector infographic poster titled "How Coffee Is Made" with five numbered steps from cherry to cup, simple icons and short labels, warm earthy palette

The original before GPT Image 2.5 Flare edited itGPT Image 2.5 Flare edit: Chalkboard to 'Today's Special: Matcha Latte', with an iced matcha latte.
GPT Image 2.5 FlareChalkboard to 'Today's Special: Matcha Latte', with an iced matcha latte.
The original before GPT Image 2.5 Flare edited itGPT Image 2.5 Flare edit: Rewrite the label as '月光' and relight it with blue moonlight.
GPT Image 2.5 FlareRewrite the label as '月光' and relight it with blue moonlight.
The original before GPT Image 2.5 Flare edited itGPT Image 2.5 Flare edit: Make it a 1998 film snapshot with an orange '98 06 21' date stamp.
GPT Image 2.5 FlareMake it a 1998 film snapshot with an orange '98 06 21' date stamp.

Settings of this model

Од
OpenAI
Тип
AI image model
Како се прави
Текст у слику · Уређивање слике
Резолуција
1K / 2K
Цена
from 90 / image

Питања

Does GPT Image 2.5 Flare spell the text I put in quotes?

In the pictures on this page it did, in English, Korean and Japanese. OpenAI says text placement and clarity can still be imperfect, so check every picture. Quote the exact words, say where they go and say whether other text may appear.

How do I stop it adding text I did not ask for?

OpenAI’s guide says to state whether extra text is allowed. In our samples the model added text when the prompt did not say. We have not tested how reliably an instruction against it works.

How should I write an edit request?

Say what to change and list what must stay, one change at a time, as OpenAI’s guide advises. In our sample edits the scene stayed recognisable around the change, but small details moved too, so look at the whole picture.

Provider documentation we read

We read these pages and wrote the points that matter in our own words. The provider may change them.

More prompting guides

Називи модела, компанија и производа на овом сајту су заштитни знаци њихових власника; Pickshot није повезан са њима, нити га они спонзоришу или подржавају.