Skip to content

Text-to-image generation benchmarks

Compare how long Qwen-Image-2.1 takes to generate a 1024×1024 image across tested local AI hardware. View first-image times and times after the model is loaded.

Qwen-Image-2.1

View image samples

Test resolution: 1024×1024. The charts show time per image after the model is loaded. First-image times in the tables also include model loading.

Base, 40 steps

Time per image

Lower is better
DGX SparkComfyUI, NVFP4
27 s
DGX SparkComfyUI, int8
29 s
Mac Studio M5 MaxComfyUI, int8, --gpu-only
1.6 min
GTR9 ProComfyUI, int8
1.6 min
Mac Studio M5 MaxComfyUI, int8
1.7 min
GTR9 ProComfyUI, NVFP4
2.8 min
MachineConfigurationPer imageFirst image
DGX SparkComfyUI, NVFP4ComfyUI, NVFP427.11 s29.36 s
DGX SparkComfyUI, int8ComfyUI, int829.39 s34.04 s
Mac Studio M5 MaxComfyUI, int8, --gpu-onlyComfyUI, int8, --gpu-only94.96 s100 s
GTR9 ProComfyUI, int8ComfyUI, int897.87 s100 s
Mac Studio M5 MaxComfyUI, int8ComfyUI, int8100 s103 s
GTR9 ProComfyUI, NVFP4ComfyUI, NVFP4171 s174 s