КАДР

GPT Image prompts: how to describe complex scenes, plus 15 ready prompts

Alen founder of KADR Updated 2026-11-04 5 min read

OpenAI's image models (GPT-5 Image Mini, GPT-5.4 Image 2 and GPT Image 2.5 Flare and Sunburst) understand long, complex descriptions better than most competitors: where each object stands, how many there are, who is doing what, which captions are needed. So you can write a detailed prompt for them, like a brief for an illustrator: point by point, with placement on the left, on the right, in the foreground. GPT-5.4 Image 2 and GPT Image 2.5 have three quality levels at different prices (on KADR (kadr-ai.ru), GPT-5.4 Image 2 low is 2 tokens, medium 6, high 17), and the top GPT Image 2.5 Sunburst on high costs 25 tokens, about 40 rubles (November 2026). The former flagship GPT-5 Image has been switched off on KADR. Below are the structure and 15 ready prompts.

When to use GPT Image

  • Many objects with a set placement.
  • Scene logic: who is doing what, who is looking at whom.
  • Diagrams, instructions, illustrations for articles.
  • Exact counts: «three cups», «five books».

For simple photorealism Flux is cheaper, for editing a finished photo use Nano Banana Pro, and for Russian text on the image Nano Banana Pro as well.

Structure

[Overall look and style]
[Object 1 + where it is]
[Object 2 + where it is]
[Action or link between them]
[Text in quotation marks, if needed]
[Format]

Write point by point. For this model a long, detailed prompt is a plus, not a minus.

Duplicate the format at the end of the prompt in the generation settings. On KADR (kadr-ai.ru), GPT-5.4 Image 2 and GPT Image 2.5 offer 16:9, 9:16, 4:3, 3:4 and others, while GPT-5 Image Mini offers only 1:1, 3:2 and 2:3.

Complex scenes

1. Office

Bright modern office illustration. On the left, a woman at a standing desk talks on the phone. In the center, two colleagues look at a laptop together. On the right, a large window with a city view and a plant. On the wall, a whiteboard with three colored sticky notes. Flat vector style, pastel colors. 16:9

2. Kitchen with an exact count

Photorealistic kitchen table seen from above: exactly three cups of coffee, one plate with two croissants, a small vase with yellow tulips in the center, a folded newspaper on the right. Soft morning light. 1:1

3. Family dinner

Warm illustration of a family dinner: grandmother at the head of the table serves a pie, a father pours tea, two children laugh, a cat sits under the table looking up. Cozy evening light, storybook style. 16:9

4. City of the future

Wide illustration of a future city street: on the left, a tram on rails covered with plants; in the center, people on bicycles; on the right, a café with an outdoor terrace; in the sky, small delivery drones. Bright daylight, detailed, optimistic mood. 16:9

Diagrams and instructions

5. A 4-step instruction

Simple instruction illustration in 4 numbered panels arranged left to right: 1) a hand opens a box, 2) takes out a coffee machine, 3) pours water into the tank, 4) presses the start button. Flat icons style, white background, numbers in circles. 16:9

6. Process diagram

Clean infographic diagram of an online order process: five connected circles from left to right with simple icons: cart, card, box, truck, house. Arrows between them. Minimal flat style, blue and white. 16:9

7. How an object is built

Exploded view illustration of a simple desk lamp: base, stem, joint, shade and bulb separated vertically with thin lines showing how they connect. Technical drawing style, white background. 3:4

Illustrations for articles

8. Remote work

Editorial illustration: a person works on a laptop on a balcony with plants, a cup of tea and a sleeping dog nearby, sunset over the city. Flat illustration, warm colors. 16:9

9. Finance

Editorial illustration: a large piggy bank in the center, small people around it add coins, one climbs a ladder, a graph with an upward arrow in the background. Flat vector, green and white. 16:9

10. Studying

Illustration of a student at a desk at night: an open laptop, a stack of three books, a desk lamp, a sticky note on the monitor, the moon in the window. Soft blue tones, flat style. 4:3

Characters and story

11. A scene with several characters

Children's book illustration: a fox gives a red scarf to a shivering rabbit in a snowy forest, an owl watches from a branch above, snowflakes fall. Soft watercolor style. 4:3

12. Comic panel

Single comic panel: a cat in a chef hat flips a pancake in a small kitchen, the pancake flies toward the ceiling, a surprised dog watches from the doorway. Bold outlines, bright flat colors. 1:1

For business

13. Post cover

Minimal cover image for a blog post about saving money: a calculator, a notebook and a cup of coffee on a light desk, top-down view, lots of empty space on the left for a title. Soft pastel colors. 16:9

14. Landing page illustration

Hero illustration for a delivery service website: a courier on an electric scooter rides through a stylized city, packages in a box behind him, a phone with a map in the corner. Flat vector, orange and white. 16:9

15. Presentation slide

Clean presentation slide illustration: three columns with simple icons: a rocket, a target and a handshake, each above an empty text area. Minimal corporate style, navy and white. 16:9

More about slides: presentation images with AI.

Which version to choose

Version Price on KADR When to use
GPT-5.4 Image 2, low 2 tokens (~3 rub.) composition drafts
GPT-5.4 Image 2, medium 6 tokens (~9 rub.) working version
GPT-5.4 Image 2, high 17 tokens (~27 rub.) final
GPT Image 2.5 Flare, low / medium / high 2 / 5 / 19 tokens (~3 / 8 / 30 rub.) fast version
GPT Image 2.5 Sunburst, low / medium / high 3 / 6 / 25 tokens (~5 / 9 / 40 rub.) details and complex scenes
GPT-5 Image Mini 7 tokens (~11 rub.) simple scenes

Prices on KADR, November 2026. This is not a public offer. Model review: GPT Image.

In short

GPT Image is strong at long, complex descriptions: object placement, counts, relationships, diagrams. Write the prompt like a brief for an illustrator, point by point, specifying «left, right, in the center». Work out the composition on GPT-5.4 Image 2 low for about 3 rubles, and make the final on high.

Frequently asked questions

How is GPT Image different from other image models?
It understands long, complex descriptions better: object placement, their number, interaction, diagrams and captions. For simple photorealism Flux is cheaper, for editing a finished photo use Nano Banana Pro.
How do I write a prompt for GPT Image?
Like a brief for an illustrator: the overall look and style, then the objects point by point with their placement (left, right, in the foreground), then the text and captions word for word in quotation marks, then the format.
What are the quality levels of GPT-5.4 Image 2?
Low, medium and high cost different amounts: 2, 6 and 17 tokens on KADR. GPT Image 2.5 Flare and Sunburst have the same quality choice. Make drafts on low and the final on high.
Does GPT Image write Russian text on images?
Better than average, but Cyrillic is weaker than Latin script in OpenAI models. For reliable Russian text Nano Banana Pro is the better choice.
How much does GPT Image cost?
On KADR, GPT-5.4 Image 2 costs from 2 to 17 tokens depending on quality (about 3-27 rubles), GPT-5 Image Mini 7 tokens (about 11 rubles), GPT Image 2.5 Sunburst from 3 to 25 tokens (about 5-40 rubles), November 2026.

Complex scenes from a description

GPT-5 Image Mini, GPT-5.4 Image 2 and GPT Image 2.5 on KADR. Drafts on low for about 3 rubles, the final on high. Pay in rubles, tokens never expire.

Open GPT Image
Alen
Founder of KADR. OpenAI's image models are connected to the service via API, and I put these prompts together around their strong side.