Skip to content

Image-editing benchmarks

Compare Qwen-Image-2.1 editing times across tested local AI hardware. Explore Base at 40 steps and Turbo at 8 steps, with results grouped by model variant.

Qwen-Image-2.1

View editing samples

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

Base, 40 steps

Time per edit

Lower is better
DGX SparkComfyUI, NVFP4
37 s
DGX SparkComfyUI, int8
41 s
Mac Studio M5 MaxComfyUI, int8, --gpu-only
1.9 min
Mac Studio M5 MaxComfyUI, int8
2.5 min
MachineConfigurationPer editFirst edit
DGX SparkComfyUI, NVFP4ComfyUI, NVFP437.25 s39.51 s
DGX SparkComfyUI, int8ComfyUI, int841.46 s42.88 s
Mac Studio M5 MaxComfyUI, int8, --gpu-onlyComfyUI, int8, --gpu-only114 s119 s
Mac Studio M5 MaxComfyUI, int8ComfyUI, int8147 s151 s

Turbo, 8 steps

Time per edit

Lower is better
DGX SparkComfyUI, int8
13 s
DGX SparkComfyUI, NVFP4
13 s
Mac Studio M5 MaxComfyUI, int8, --gpu-only
30 s
GTR9 ProComfyUI, int8
47 s
Mac Studio M5 MaxComfyUI, int8
63 s
GTR9 ProComfyUI, NVFP4
64 s
MachineConfigurationPer editFirst edit
DGX SparkComfyUI, int8ComfyUI, int813.00 s15.55 s
DGX SparkComfyUI, NVFP4ComfyUI, NVFP413.11 s15.82 s
Mac Studio M5 MaxComfyUI, int8, --gpu-onlyComfyUI, int8, --gpu-only30.19 s36.12 s
GTR9 ProComfyUI, int8ComfyUI, int847.01 s51.60 s
Mac Studio M5 MaxComfyUI, int8ComfyUI, int862.64 s65.89 s
GTR9 ProComfyUI, NVFP4ComfyUI, NVFP464.40 s69.41 s