Skip to content
ModelsThatRun

Text to image: General

General-purpose text-to-image models.

1251020501000.40.50.60.70.80.9Time per image on RTX 4090, s, log scale (measured)GenEval (measured)
Commercial OKConditionalNon-commercialBest trade-off so far1 not yet scored for accuracy, not plottedInput sizes differ: compare latency with care

All general models

Each model runs at its official resolution and number of steps, shown under its time. Fewer steps and smaller images are less work, so compare times with care.

General models, best measured accuracy first
ModelAccuracy (paper under it)4090 per image (size and steps under it)VRAM peakLicense
FLUX.2 [klein]2 sizes: 4b, base-4b
0.82 GenEval
1.30 s1024x1024 · 4 steps
20.0 GiB
Commercial OK
Sana 1.52 sizes: 1.6b, 4.8b
0.81 GenEval0.82 GenEval overall paper
1.53 s1024x1024 · 20 steps
12.1 GiB
Conditional
Z-Image2 sizes: turbo, base
0.76 GenEval0.82 GenEval overall paper
4.50 s1024x1024 · 9 steps
22.7 GiB
Commercial OK
Sana-Sprint1 size: 1.6b
0.75 GenEval0.77 GenEval overall paper
0.27 s1024x1024 · 2 steps
12.0 GiB
Conditional
Lumina-Image 2.01 size: 2.0
0.63 GenEval0.73 GenEval overall paper
13.7 s1024x1024 · 50 steps
15.3 GiB
Conditional
Playground v2.51 size: 1024px-aesthetic
0.59 GenEval
5.99 s1024x1024 · 50 steps
14.6 GiB
Conditional
PixArt-Σ1 size: xl-2-1024-ms
0.55 GenEval
2.47 s1024x1024 · 20 steps
16.9 GiB
Conditional
SDXL-Turbo1 size: 1.0
0.55 GenEval
0.15 s512x512 · 1 step
9.3 GiB
Conditional
Stable Diffusion XL1 size: base
0.54 GenEval
5.97 s1024x1024 · 50 steps
14.6 GiB
Conditional
SDXL-Lightning1 size: 4step
0.54 GenEval
0.56 s1024x1024 · 4 steps
14.4 GiB
Conditional
Stable Diffusion 1.51 size: v1-5
0.43 GenEval
1.39 s512x512 · 50 steps
4.2 GiB
Conditional