AI images with GPT Image — capabilities, modes and uses

What GPT Image is

GPT Image is the family of image generation models developed by OpenAI — the company behind ChatGPT and the DALL·E series. It is a multimodal model that turns a text description into an image and can edit an already existing photo according to an instruction in plain language. GPT Image builds on a language model, which helps it understand more descriptive and more complex requests.

The model became widely known through image generation in ChatGPT, where the user describes an idea in words and gets a finished visual. Its strong side is the link between language and picture: it does well with layered instructions, with the arrangement of the elements in the frame and — something that was a weak spot of AI generators for a long time — with spelling out legible text on the image itself.

On generiram, GPT Image is available through the AI Studio and covers two main scenarios: creating an image from text and editing an uploaded photo. That way one and the same model does the job both for completely new visuals and for precise changes to existing material.

Versions and options on generiram

At the moment GPT Image 2 is available on generiram — the current generation of the model — in two working modes:

VariantModeWhat it does
GPT Image 2Text to imageCreates a new image from a text description (a prompt) alone. A good fit when you are starting from an idea rather than from a finished file.
GPT Image 2Image editingTakes an uploaded photo and changes it according to a text instruction — adding, removing or altering elements, style or details, while keeping the rest of the frame.

Text to image

In this mode you describe the scene you want in words, and the model builds it from scratch. The clearer the description — object, surroundings, style, mood, composition — the closer the result is to the idea. The mode is handy for getting from concept to a finished visual quickly, with no source material needed.

Image editing

Here you upload an existing image and say in text what should change. The model tries to apply the change while keeping the context of the photo — the surroundings, the light and the composition. This suits targeted corrections on a particular shot, rather than generating an entirely new image.

Strengths

  • Legible text in the image. GPT Image is among the models that do well at spelling out captions, labels and short texts on the visual — often a weak spot for other generators.
  • Understanding complex instructions. Thanks to its language foundation the model follows more descriptive and layered prompts, including the arrangement of and the relationships between several elements in the frame.
  • Broad knowledge of the world. The model recognises many objects, styles and concepts, which makes it easier to describe scenes in everyday words.
  • Consistent editing. When changing an existing photo, it aims to keep the rest of the frame rather than rewriting the whole image.
  • Varied styles. From a photorealistic look to illustration and graphics — one and the same model covers a wide spectrum of visual styles.

What it is good for

  • Marketing and social media — visuals for ads, banners, posts and covers, including ones with text on the image.
  • Product and concept images — quick visualisations of ideas, products and scenes before the real production.
  • Illustrations and posters — where combining image and caption matters.
  • Photo editing — targeted changes, adding or removing elements and altering details on a particular shot.
  • Prototypes and mood board ideas — generating variants for direction and style at the start of a project.

How to start

GPT Image 2 is used through the genkiki.com AI Studio. Pick the GPT Image 2 model and the mode you need — text to image for a new image or image editing to change an existing one. Describe clearly what you want (and for editing, upload the photo you will work on) and start the generation. If the result is not exactly what you want, refine the description or the instruction and try again — a more specific prompt usually gives a result closer to the idea.

The GPT Image models — in detail

Which one is good for what, where it falls short and what it shouldn't be used for.

GPT Image 2

photophoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 441 credits

Strong at

  • Edits a photo with the same LETTER accuracy — about 99% correct typography, foreign scripts included
  • Accepts up to 16 reference photos in one request
  • Reasons before it touches anything — so it gets “change only this, leave the rest”
  • Native 2048-pixel output
  • Gives up to 8 edit variations in a consistent style

Weak at

  • Strict filter — refuses edits with real people and brands
  • Smooths out the style: the photo's original “grain” gets lost
  • Expensive with many outputs

Don't use it for

  • Don't use it to edit a photo of a recognizable celebrity
  • Don't use it when the exact texture of the original must be kept
  • Don't use it for ten tests in a row

Typical tasks

  • Change the text in a finished ad
  • Add or remove an element precisely
  • Make several same-style variations of one image
  • Editing with several reference photos

GPT Image 2.5 Flare

photophoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 106 credits

Strong at

  • Разбира най-добре какво точно да се промени в подадената снимка
  • До 16 входни снимки наведнъж
  • Приема и маска — посочваш коя част да се пипне, останалото остава
  • Шест степени на детайлност
  • Пише вярно текст на кирилица: надписи, табели и слогани на български

Weak at

  • Stricter filter — refuses topics that other models let through
  • The highest levels take more time and cost more
  • For clean typography, the specialized models are more accurate

Don't use it for

  • Don't use it for a simple background swap — there are faster ones

Typical tasks

  • Смени само посочената част от снимката
  • Слей няколко снимки в един кадър
  • Многостъпкова редакция по дълго описание
  • Постер, банер или етикет с надпис на български

GPT Image 2.5 Sunburst

photophoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 106 credits

Strong at

  • Най-внимателната редакция в семейството
  • До 16 входни снимки наведнъж
  • Приема и маска — посочваш коя част да се пипне
  • Шест степени на детайлност
  • Пише вярно текст на кирилица: надписи, табели и слогани на български

Weak at

  • Stricter filter — refuses topics that other models let through
  • The highest levels take longer and cost more
  • For clean typography, specialized models are more accurate
  • Slower than Flare for the same task

Don't use it for

  • Don't use it for a small fix that won't be viewed up close

Typical tasks

  • Фина поправка, която ще се разглежда отблизо
  • Редакция на снимка за печат
  • Сложна промяна с маска върху голяма снимка
  • Постер, банер или етикет с надпис на български

GPT Image 2

textphoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 425 credits

Strong at

  • The most accurate with LETTERS: the maker claims about 99% correct typography, and over 95% in foreign scripts (Chinese, Japanese, Korean, Hindi, Bengali, Arabic)
  • Native 2048 pixels, with larger sizes in testing — higher than most big models
  • Accepts up to 16 reference photos in one request — the most among the leaders
  • The first with built-in REASONING: it thinks over the prompt before drawing, so it handles complex conditions
  • Fast for its tier — about 3 seconds, three to five times faster than its predecessor
  • Gives up to 8 images in a consistent style from one prompt

Weak at

  • Strict filter: refuses prompts that other models let through — especially with real people and brands
  • Stylistically “cleaner” and more impersonal than the artistic models
  • 4K is still in testing, not guaranteed
  • Expensive for many outputs at once

Don't use it for

  • Don't use it when the prompt is on the edge of the filter — you'll waste time on refusals
  • Don't use it for a distinctly artistic or “gritty” style — it comes out too polished
  • Don't count on 4K as a sure thing — it's still in testing

Typical tasks

  • Poster, packaging or label with lots of exact text
  • Text in Bulgarian or another script inside the image itself
  • Complex scene with conditions described in detail
  • A series of images in a consistent style for a campaign
  • Composition from several reference photos

GPT Image 2.5 Flare

textphoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 106 credits

Strong at

  • Най-послушният в каталога, с шест степени на детайлност вместо три
  • Разбира сложни описания с много условия наведнъж
  • По-бърз от Sunburst при същото качество
  • Изход до 3840 пиксела по дългата страна
  • Пише вярно текст на кирилица: надписи, табели и слогани на български

Weak at

  • Stricter filter — refuses topics that other models let through
  • The highest levels take more time and cost more
  • For clean typography, the specialized models are more accurate

Don't use it for

  • Don't use it at the highest levels for a quick series of tests — there the low level gives the same shot for a fraction of the price

Typical tasks

  • Сложна сцена с много условия
  • Всекидневна работа, където бързината има значение
  • Проба на ниска степен, после същият кадър на висока
  • Постер, банер или етикет с надпис на български

GPT Image 2.5 Sunburst

textphoto
Aspect ratios: 1:1, 9:16, 16:9 Price: 106 credits

Strong at

  • Същото послушание като Flare, но с повече внимание към детайла
  • Шест степени на детайлност вместо три
  • Най-добрият избор, когато кадърът ще се гледа отблизо
  • Изход до 3840 пиксела по дългата страна
  • Пише вярно текст на кирилица: надписи, табели и слогани на български

Weak at

  • Stricter filter — refuses topics that other models let through
  • The highest levels take longer and cost more
  • For clean typography, specialized models are more accurate
  • Slower than Flare for the same task

Don't use it for

  • Don't use it for quick tests — for those, Flare gives the same in less time

Typical tasks

  • Кадър за печат или за голям екран
  • Сцена с фина текстура и дребни подробности
  • Финалната снимка, след като пробите са минали през Flare
  • Постер, банер или етикет с надпис на български

Frequently asked questions about GPT Image

How much does one generation with GPT Image cost?

You pay with credits and the price depends on the model: GPT Image 2 — 441 credits; GPT Image 2.5 Flare — 106 credits; GPT Image 2.5 Sunburst — 106 credits; GPT Image 2 — 425 credits; GPT Image 2.5 Flare — 106 credits; GPT Image 2.5 Sunburst — 106 credits. The exact price is shown before you start the generation.

What are the limitations of GPT Image?

Here is what each model supports: GPT Image 2 — aspect ratios 1:1, 9:16, 16:9; GPT Image 2.5 Flare — aspect ratios 1:1, 9:16, 16:9; GPT Image 2.5 Sunburst — aspect ratios 1:1, 9:16, 16:9; GPT Image 2 — aspect ratios 1:1, 9:16, 16:9; GPT Image 2.5 Flare — aspect ratios 1:1, 9:16, 16:9; GPT Image 2.5 Sunburst — aspect ratios 1:1, 9:16, 16:9.

What is GPT Image best for?

Strengths by model: GPT Image 2 — Edits a photo with the same LETTER accuracy — about 99% correct typography, foreign scripts included, Accepts up to 16 reference photos in one request, Reasons before it touches anything — so it gets “change only this, leave the rest”; GPT Image 2.5 Flare — Разбира най-добре какво точно да се промени в подадената снимка, До 16 входни снимки наведнъж, Приема и маска — посочваш коя част да се пипне, останалото остава; GPT Image 2.5 Sunburst — Най-внимателната редакция в семейството, До 16 входни снимки наведнъж, Приема и маска — посочваш коя част да се пипне.

GPT Image vs the other models

Head to head: price in credits, maximum length, resolutions — and where each of the two falls short.

GPT Image vs Nano Banana Nano Banana vs GPT Image: how many images they accept at once for editing, aspect ratios, resolutions and pri… GPT Image vs Imagen GPT Image vs Imagen 4: one model with editing versus three variants for generation only. Price in credits, ti…

All model comparisons

Try GPT Image now

Generate your own images with GPT Image right in the browser — no installation, in English.

Open the AI Studio See the prices