Which AI image model follows a hard composition

This prompt is a counting and placement test: a blue book on the left, a glass of water with three ice cubes in the middle, a red apple on the right, a pencil balanced on the glass. We asked 23 models and checked every position. The grid includes the September 18 additions; each model page dates its own run.

Anvisha Pai

Anvisha Pai, Co-founder & CEO, Voyager

Verified

What the side-by-side shows

GPT Image 2.5 Flare, Nano Banana Pro, FLUX.2 Pro and Grok Imagine 2.0 placed everything exactly, pencil balanced and all. Most of the roster is close. Recraft V3 put the pencil on the table and the apple behind the glass; Recraft V4.1 stood the pencil inside the glass; Ideogram 4.0 and Seedream 5.0 Lite lost the count of ice cubes.

  1. 1.GPT Image 2.5 Flare every object in its place, pencil balanced on the rim.
  2. 2.Nano Banana Pro exact placement with window light.
  3. 3.FLUX.2 Pro correct order and count, cleanly lit.
  4. 4.Grok Imagine 2.0 three cubes, pencil on the glass, apple on the right.

Every image model on this prompt, 23 pictures

One try each, nothing re-rolled or retouched, generated by us. Ranked models come first; the rest follow in family order. Open a model for the rest of its set.

›The prompt every model got

A glass of water on a wooden table with exactly three ice cubes in it, a yellow pencil resting across the rim of the glass, a blue book to the left of the glass and a red apple to the right, photographed from table height.

Exact counts and left-right placement of five named objects.

GPT Image 2.5 Flare result for the hard composition prompt
1GPT Image 2.5 Flare21s · $0.04116

every object in its place, pencil balanced on the rim.

Nano Banana Pro result for the hard composition prompt
2Nano Banana Pro18s · $0.15

exact placement with window light.

FLUX.2 Pro result for the hard composition prompt
3FLUX.2 Pro13s · $0.045

correct order and count, cleanly lit.

Grok Imagine 2.0 result for the hard composition prompt
4Grok Imagine 2.080s · $0.06

three cubes, pencil on the glass, apple on the right.

FLUX.2 Flex result for the hard composition prompt
FLUX.2 Flex12s · $0.10
FLUX.2 Max result for the hard composition prompt
FLUX.2 Max29s · $0.10
GPT Image 2.5 Sunburst result for the hard composition prompt
GPT Image 2.5 Sunburst31s · $0.04116
GPT Image 2 result for the hard composition prompt
GPT Image 2103s · $0.165
Grok Imagine Pro result for the hard composition prompt
Grok Imagine Pro8s · $0.05
Grok Imagine result for the hard composition prompt
Grok Imagine7s · $0.02
HiDream-I1 Full result for the hard composition prompt
HiDream-I1 Full13s · $0.10
Ideogram 4.0 result for the hard composition prompt
Ideogram 4.013s · $0.0209
Krea 2 Medium result for the hard composition prompt
Krea 2 Medium16s · $0.03
Krea 2 Large result for the hard composition prompt
Krea 2 Large31s · $0.06
Nano Banana 2 result for the hard composition prompt
Nano Banana 213s · $0.08
Nano Banana 2 Lite result for the hard composition prompt
Nano Banana 2 Lite3s · $0.034
Qwen Image 3 result for the hard composition prompt
Qwen Image 323s · $0.075
Qwen-Image result for the hard composition prompt
Qwen-Image7s · $0.04
Recraft V4.1 result for the hard composition prompt
Recraft V4.18s · $0.035
Recraft V3 result for the hard composition prompt
Recraft V36s · $0.04
Seedream 5.0 Pro result for the hard composition prompt
Seedream 5.0 Pro34s · $0.0675
Seedream 5.0 Lite result for the hard composition prompt
Seedream 5.0 Lite32s · $0.035
Seedream 4.5 result for the hard composition prompt
Seedream 4.510s · $0.04

The other prompts

Frequently asked questions

Which AI image generator follows prompts most precisely?

On a five-object placement test, GPT Image 2.5 Flare, Nano Banana Pro, FLUX.2 Pro and Grok Imagine 2.0 got every position right. The Recraft models were the loosest.

Anvisha Pai

Anvisha Pai

Co-founder & CEO, Voyager

Anvisha is the CEO of Voyager and a repeat, Y Combinator-backed startup founder. She was previously a PM at Dropbox. She believes nobody should need a design degree to make something that looks great.

Voyager

Run any of these inside Voyager

Voyager is an agent for creative work that runs every model on this page. Brief it, and it picks the model for the job, makes the thing and hands back something you can edit. Create an account to get started.