Text rendering benchmark

Best AI image generator for text rendering

Text in generated images is still a practical failure point: spelling, small labels, poster copy, typography, and layout all matter. ImageBench V1 ranks models on the same text-rendering prompts so the comparison is evidence-based.

Current leader: openai/gpt-image-2 at 100.0%.

openai/gpt-image-2 text rendering benchmark example — openai/gpt-image-2
100.0% text score

fal/google/nano-banana-2 text rendering benchmark example — fal/google/nano-banana-2
86.7% text score

bfl/flux-2-max text rendering benchmark example — bfl/flux-2-max
100.0% text score

bfl/flux-2-pro text rendering benchmark example — bfl/flux-2-pro
80.0% text score

Text rendering ranking

This score covers typography style, spelling accuracy, multi-line copy, posters, product labels, and small readable text.

Rank	Model	Text rendering	Overall	Pass / Fail
1	openai/gpt-image-2	100.0%	96.4%	185 / 7
2	bfl/flux-2-max	100.0%	91.7%	176 / 16
3	fal/google/nano-banana-pro	100.0%	91.1%	175 / 17
4	fal/google/nano-banana-2	86.7%	95.3%	183 / 9
5	bfl/flux-2-pro	80.0%	83.3%	160 / 32
6	z-image-local/z-image-turbo	73.3%	75.5%	145 / 47
7	qwen-image-local/qwen-image-gen	60.0%	70.8%	136 / 56
8	nucleus-local/nucleus-image	60.0%	67.2%	129 / 63
9	bfl/flux-2-klein-9b	40.0%	78.6%	151 / 41
10	bfl/flux-2-klein-4b	40.0%	74.0%	142 / 50
11	sana-local/sana-1.5-1.6b	20.0%	53.1%	102 / 90

Examples from the benchmark

openai/gpt-image-2

100.0% text score, 15/15 pass

Details

A poster with headline 'SUMMER SALE' in large bold text, body text 'Up to 50% off all items this weekend only' in medium size, and fine print 'Terms and conditions apply. See store for details.' at the bottom in small text

PASS