Head to head
Esc

Models · slot A

to navigate to pick

Geometric Composition

Vote

A spatial-reasoning test. Each object has a precise relationship (inside, on top, behind, seen through the glass), so it measures whether a model follows explicit placement instructions and handles transparency and refraction rather than just approximating the scene.

Voters judged on
Spatial Accuracy Object Relationships Glass & Refraction Lighting Direction
The brief Every model got the same prompt
Prompt
A glass cube on a wooden table. Inside the cube is a small blue sphere. On top of the cube sits a red book. A green plant is behind the cube, partially visible through the glass. Soft window light from the left.
The leaderboard

Challenge rankings

60 models · ranked by blind vote

Through time: how it evolved
1
ImagineArt 1.5 (Preview)
score 28.9
-1
2
FLUX.2 [dev] Flash
score 27.7
0
3
Seedream 4.5
score 27.3
0
4
Nano Banana 2
score 26.4
0
5
FLUX.2 [dev] Turbo
score 26.1
0
6
Nano Banana 2 Lite
score 24.1
0
7
GPT Image 1 Mini
score 24.0
0
8
Wan 2.5 (Preview)
score 23.7
0
9
Nano Banana Pro
score 23.4
1
10
GPT Image 1.5
score 22.5
0
11
Seedream 4.0
score 22.4
0
12
FLUX.2 [max]
score 22.3
0
13
FLUX.2 [dev]
score 21.3
0
14
FLUX.2 [flex]
score 21.0
0
15
Grok Imagine Image Pro
score 20.6
0
16
Z-Image Turbo
score 20.5
0
17
Nano Banana
score 19.7
0
18
HiDream I1 Full
score 18.8
0
19
FLUX.2 [pro]
score 18.4
0
20
Imagen 4.0 Ultra Generate 001
score 18.3
0
21
Seedream 5.0 Lite
score 18.2
0
22
Recraft V4.1 Pro
score 18.2
0
23
FLUX.2 [klein] 9B
score 18.0
0
24
Qwen Image 2512
score 18.0
0
25
Grok Imagine Image
score 17.8
0
26
FLUX.1 Kontext [max]
score 17.6
0
27
Recraft V4
score 17.4
0
28
GPT Image 1
score 17.3
0
29
Vidu Q2
score 17.3
0
30
Recraft V4 Pro
score 16.7
0
31
Stable Diffusion 3.5 Large
score 16.4
0
32
Wan 2.7 Pro
score 16.0
0
33
Reve Image 1.0
score 15.9
0
34
Wan 2.6
score 14.8
0
35
FLUX.1 [schnell]
score 14.4
0
36
FLUX.2 [klein] 4B
score 13.8
0
37
FLUX.1 [schnell] FP8
score 13.2
0
38
FLUX1.1 [pro]
score 12.6
0
39
Qwen Image 2.0
score 12.4
0
40
FLUX.1 Kontext [dev]
score 12.0
0
41
Bria FIBO
score 11.7
0
42
FLUX.1 Krea [dev]
score 11.4
0
43
Stable Diffusion 3.5 Medium
score 10.4
0
44
Stable Diffusion 3.5 Large Turbo
score 10.4
0
45
DALL-E 2
score 10.2
0
46
DALL-E 3
score 9.8
0
47
Qwen Image
score 8.5
0
48
LongCat-Image
score 8.1
0
49
Wan 2.7
score 7.9
0
50
GPT Image 2
score 7.9
0
51
Qwen Image 2.0 Pro
score 7.0
0
52
FLUX.1 [dev]
score 6.9
0
53
Recraft V4.1
score 6.6
0
54
FLUX1.1 [pro] Ultra
score 6.2
0
55
FLUX.1 Kontext [pro]
score 1.0
0
56
OmniGen v2
score 0.9
0
57
Qwen Image Max
score 0.3
0
58
HiDream I1 Fast
score 0.0
0
Imagen 4.0 Generate 001
awaiting votes
0
Imagen 4.0 Fast Generate 001
awaiting votes
0
Head to head 4 matchups

Notable battles

The matchups worth a second look from this challenge's blind voting: the closest rivalries, the biggest upsets (a lower seed taking down a favorite), and the clashes at the top of the board. Each bar shows how the community split its votes.

Want to compare other models? Pick any two and see them go head to head.
The archive

Through time

How Geometric Composition has evolved, release by release.

2022 to 2026· 10 reigns· 60 models
The Reign Chain
10 reigns

How the crown changed hands

Every model that has held #1 on this challenge, in order. Click any era to travel through time.

  1. R01
    Inception
    DALL-E 2
    Apr 2022 Jul 2024
    reigned 2.3 yrs
  2. R02
    model
    Jul 2024 Aug 2024
    reigned 1 mo · overtook DALL-E 2
  3. R03
    FLUX.1 [schnell]
    Aug 2024 Oct 2024
    reigned 2 mo
  4. R04
    Stable Diffusion 3.5 Large
    Oct 2024 Mar 2025
    reigned 5 mo · overtook FLUX.1 [schnell]
  5. R05
    GPT Image 1
    Mar 2025 Apr 2025
    reigned 1 mo · overtook Stable Diffusion 3.5 Large
  6. R06
    HiDream I1 Full
    Apr 2025 Sep 2025
    reigned 5 mo · overtook GPT Image 1
  7. R07
    Seedream 4.0
    Sep 2025 Sep 2025
    reigned 1 mo · overtook HiDream I1 Full
  8. R08
    Wan 2.5 (Preview)
    Sep 2025 Oct 2025
    reigned 1 mo · overtook Seedream 4.0
  9. R09
    GPT Image 1 Mini
    Oct 2025 Nov 2025
    reigned 1 mo · overtook Wan 2.5 (Preview)
  10. R10
    Current
    ImagineArt 1.5 (Preview)
    Nov 2025 today
    reigned 10 mo · overtook GPT Image 1 Mini

Generation history

60 models · newest first
Show full history 55
  1. Recraft V4.1 3 takes
  2. GPT Image 2 3 takes
  3. Wan 2.7 3 takes
  4. 55 more models Apr 2022 – Apr 2026 Show