Which AI model fools people most?

Live ranking from real quiz answers. Genuine photographs are fixed at 1500 — anything above that line gets picked as “the real one” more often than reality does.

Limited to the models built into this quiz — not every model on the market, and not always the latest version of each. A model's absence here says nothing about how good it is.

Image models

31 models listed · 846 AI images in the quiz · click a column to sort

			Confidence
—	Real image	1500	—	46994	28676	61%	—
1	Ideogram 4.0 QualityIdeogram	1387	1365–1408	1296	335	26%	10
2	Nano Banana 2 with Web SearchGoogle	1381	1354–1408	859	219	25%	10
3	Recraft V3Recraft	1370	1345–1395	944	252	27%	10
4	Imagen 2Google	1357	1344–1370	6934	1340	46%	29
5	Grok Imagine Image QualityxAI	1353	1327–1380	951	213	22%	10
6	Nano Banana 2 LiteGoogle	1343	1318–1367	1162	247	21%	10
7	Qwen Image 2.0 ProAlibaba	1325	1301–1350	1172	236	20%	10
8	Reve 2.1Reve AI	1324	1300–1347	1325	271	20%	10
9	Z-ImageAlibaba	1287	1280–1294	17307	3010	18%	132
10	Kling Image V3Kuaishou	1284	1262–1307	1650	281	17%	10
11	Juggernaut Pro FluxRunDiffusion	1284	1272–1296	6289	1053	20%	32
12	FLUX.2 [klein] 9BBlack Forest Labs	1257	1228–1286	1238	184	19%	14
13	Stable Diffusion 3.5 LargeStability AI	1255	1236–1274	2433	375	15%	10
14	Seedream 4.5ByteDance	1248	1237–1258	8418	1233	15%	33
15	Imagen 3 FastGoogle	1243	1181–1305	237	35	15%	10
16	FLUX.2 [pro]Black Forest Labs	1235	1202–1268	957	123	13%	10
17	Stable DiffusionStability AI	1232	1217–1248	4643	601	18%	37
18	Gemini 2.5 Flash Image / Nano BananaGoogle	1231	1214–1247	3784	510	13%	14
19	FLUX.2 [max]Black Forest Labs	1212	1188–1236	1936	231	12%	10
20	Imagen 4 UltraGoogle	1201	1164–1239	860	93	11%	10
21	FLUX.1 [schnell]Black Forest Labs	1183	1175–1191	21726	2130	10%	157
22	Kling IMAGE 3.0 OmniKuaishou	1182	1164–1200	4033	410	10%	24
23	GPT Image 1.5OpenAI	1176	1161–1191	5729	564	10%	23
24	SDXL-LightningByteDance	1172	1158–1185	7969	720	9%	72
25	FLUX.1 [dev]Black Forest Labs	1166	1153–1180	8237	748	9%	41
26	MAI-Image-2.5Microsoft AI	1166	1122–1211	730	64	9%	10
27	Juggernaut Flux LightningRunDiffusion	1151	1122–1179	1794	159	9%	10
28	DALL·E 3OpenAI	1148	1122–1173	2634	195	7%	27
29	Seedream 4ByteDance	1138	1114–1162	2634	215	8%	13
30	Seedream 5.0 ProByteDance	1090	1006–1174	276	17	6%	10
31	Qwen-Image-MaxAlibaba	1084	1051–1117	1842	113	6%	12

Dimmed rows have fewer than 60 appearances — early evidence, not yet a settled rating.

A model joins this table once it has at least 10 images in the quiz and has appeared in at least 30 answers — below that, the number would say more about one lucky (or unlucky) image than about the model itself. New models and images are added regularly, so more entries — and firmer ratings — are on the way.

In the quiz but not yet listed: FLUX.2 LoRA Gallery: Realism (fal), FLUX1.1 [pro] (Black Forest Labs), GPT Image 2 Medium (OpenAI), Imagen 4 (Google), Lucid Realism (Leonardo AI), Titan Image Generator G1 v1 (Amazon) — fewer than 10 images in the quiz so far.

Last recomputed 18 Sept 2026, 03:15 · refreshed nightly

Add your own answers

Every round you play feeds straight into these numbers.

Architecture and geometry

Paintings

Brushwork and composition

Cars

Reflections and badges

🎬 Video quizzes

Not in the ranking above yet — too few answers so far to rate video models.

Gravity, fire, liquids

How the rating works

Bradley-Terry, not a win count. Each round counts as a choice among the options actually on screen, so beating strong rivals is worth more than beating weak ones. Two-option rounds and four-option rounds therefore combine without their different odds distorting anything.

Read the range, not the position. Where two intervals overlap, the order between those models is not established. Dimmed rows have fewer than 20 appearances and are listed for completeness only.

Pool size is context. A model represented by three images is judged on those three images, however often they were shown.

Replays don't get extra weight — within a visit. Only your first look at a round counts, because after that you'd recognise it and it stops being a fresh judgment. That rule ends when you close the tab: come back later the same day and a repeat is counted again.

Limits. The method assumes all players judge alike, and this measures our image pool rather than a model's ceiling. Answers recorded before 29 August 2026 come only from players who accepted analytics cookies; since then every player counts, so figures spanning that date mix two different groups.