Which AI image model renders text best

Text in an image was the classic failure of image models and is now the clearest way to tell them apart. We asked 36 of them for the same poster: a title, a date line and a footer, in a specified style, and kept every result. 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

The OpenAI models lead: GPT Image 2.5 Flare and Sunburst and GPT Image 2 set all three lines correctly with real poster hierarchy. Nano Banana Pro is as good, with the boldest type on the roster, and Seedream 5.0 Pro and Grok Imagine Pro are clean. FLUX.2 Max and Pro spell correctly but with a slightly softer finish. The failures: Seedream 4.5 broke the layout and dropped a stray 'D', Recraft V3 printed '@pm', Recraft V4.1 turned the words into unreadable shapes, Ideogram 4.0 left most of the poster empty, FLUX.2 Flex overlapped its lines, and Grok Imagine 1.0 let the words collide. Of the two open-weight additions, Qwen-Image set the poster cleanly and HiDream-I1 overlapped its title lines. FLUX 3 Image, added in October, set all three lines correctly in a two-colour layout with a clear hierarchy. Ideogram 4.5, added on 22 October, set all three lines correctly at every quality tier, with an extra 'LONDON' tag it was not asked for. Of the nine models added on 22 October, Krea 3, FLUX 3 Image Klein, Seedream 5.0 Flash, Muse Image, Bria FIBO 2.0, MOSS-Image and Recraft V4.1 Flash all set the three lines correctly with nothing added; both MAI-Image-2.5 models spelled them right but set the smaller lines in capitals. Ideogram Vector preserves the requested wording but adds an unwanted SUD IMERA line above it.

  1. 1.GPT Image 2.5 Flare three lines, correct, with real poster hierarchy and a scene behind them.
  2. 2.Nano Banana Pro the boldest, cleanest type on the roster.
  3. 3.GPT Image 2 correct text and a complete poster from the previous OpenAI model.
  4. 4.Seedream 5.0 Pro clean lettering with a bit of graphic flair.
  5. 5.Grok Imagine Pro every word right, in a busier layout.

Every image model on this prompt, 36 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 concert poster that reads 'MODA SUMMER SESSIONS' in large bold type, with the subtitle 'Friday 12 June, 8pm, Roundhouse' and a smaller line 'Tickets at the door'. Risograph style, two colours, tangerine and navy, grainy ink texture.

Spelling, kerning and hierarchy of three lines of set text.

GPT Image 2.5 Flare result for the text in an image prompt

three lines, correct, with real poster hierarchy and a scene behind them.

Nano Banana Pro result for the text in an image prompt
2Nano Banana Pro37s · $0.15

the boldest, cleanest type on the roster.

GPT Image 2 result for the text in an image prompt
3GPT Image 2117s · $0.17

correct text and a complete poster from the previous OpenAI model.

Seedream 5.0 Pro result for the text in an image prompt
4Seedream 5.0 Pro56s · $0.068

clean lettering with a bit of graphic flair.

Grok Imagine Pro result for the text in an image prompt

every word right, in a busier layout.

Bria FIBO 2.0 result for the text in an image prompt
Bria FIBO 2.019s · $0.04
FLUX 3 Image result for the text in an image prompt
FLUX 3 Image24s · $0.024
FLUX 3 Image Klein result for the text in an image prompt
FLUX 3 Image Klein3s · $0.009
FLUX.2 Pro result for the text in an image prompt
FLUX.2 Pro10s · $0.045
FLUX.2 Flex result for the text in an image prompt
FLUX.2 Flex11s · $0.10
FLUX.2 Max result for the text in an image prompt
FLUX.2 Max25s · $0.10
GPT Image 2.5 Sunburst result for the text in an image prompt
Grok Imagine 2.0 result for the text in an image prompt
Grok Imagine 2.060s · $0.06
Grok Imagine result for the text in an image prompt
Grok Imagine32s · $0.02
HiDream-I1 Full result for the text in an image prompt
HiDream-I1 Full104s · $0.10
Ideogram 4.5 result for the text in an image prompt
Ideogram 4.512s · $0.06
Ideogram 4.0 result for the text in an image prompt
Ideogram 4.013s · $0.021
Ideogram Vector result for the text in an image prompt
Ideogram Vector25s · $0.16
Krea 3 result for the text in an image prompt
Krea 347s · $0.075
Krea 2 Medium result for the text in an image prompt
Krea 2 Medium25s · $0.03
Krea 2 Large result for the text in an image prompt
Krea 2 Large47s · $0.06
MAI-Image-2.5 result for the text in an image prompt
MAI-Image-2.533s · $0.05
MAI-Image-2.5 Pro result for the text in an image prompt
MAI-Image-2.5 Pro27s · $0.17
Muse Image result for the text in an image prompt
Muse Image15s · $0.01
MOSS-Image result for the text in an image prompt
MOSS-Image20s · $0.066
Nano Banana 2.1 result for the text in an image prompt
Nano Banana 2.116s · $0.04
Nano Banana 2 result for the text in an image prompt
Nano Banana 213s · $0.08
Nano Banana 2 Lite result for the text in an image prompt
Nano Banana 2 Lite4s · $0.034
Qwen Image 3 result for the text in an image prompt
Qwen Image 321s · $0.075
Qwen-Image result for the text in an image prompt
Qwen-Image8s · $0.04
Recraft V4.1 result for the text in an image prompt
Recraft V4.116s · $0.035
Recraft V4.1 Flash result for the text in an image prompt
Recraft V4.1 Flash14s · $0.007
Recraft V3 result for the text in an image prompt
Recraft V37s · $0.04
Seedream 5.0 Lite result for the text in an image prompt
Seedream 5.0 Lite38s · $0.035
Seedream 5.0 Flash result for the text in an image prompt
Seedream 5.0 Flash20s · $0.027
Seedream 4.5 result for the text in an image prompt
Seedream 4.516s · $0.04

The other prompts

Frequently asked questions

Which AI image generator is best for text?

On a three-line poster, GPT Image 2.5, GPT Image 2, Nano Banana Pro, Seedream 5.0 Pro and Grok Imagine Pro all set every word correctly. Ideogram, which markets text rendering, left most of the poster blank on this prompt.

Can Midjourney render text?

Midjourney is not on this roster. Of the models here, the OpenAI and Google models were the most reliable with words.

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.