⚡ Local Inference Recipes

English | 简体中文 | 日本語

Qwen-Image: local inference speed (tok/s), sourced

As of 2026-11-14, this dataset has 13 recipes for Qwen-Image. 7 were measured by the poster. These are image, video, or audio results. Speed is in the metric column, not tok/s. 2 entries have a date, from 2026-10-11 to 2026-10-30.

Filter these in the explorer →

ModelEngineHardwaretok/sMetricSource levelDateSource
Qwen-Image 2.1 (7B + 8B text encoder)ComfyUIRTX 4090 24GB8.1 s / 1024² (40 steps); 30.2 s / 2048×1152 (50 steps); peak 16.2GBMeasuredlink
Qwen-Image 2.1TensorFoldRTX 5090 @450W4.53 s / 6.94 s / 13.87 s (1024², 25 steps; NVFP4 peak 22.9GB; images not identical to Q6)Measuredlink
Qwen-Image 2.1UnspecifiedRTX 5090 32GB (455W power limit)Qwen-Image 2.1 base model, int8, 25 steps, 1MP: 5.2 s per image (median of the last 200 images), about 687 images per hour; power 399 W (nvidia-smi 60 s average); energy 0.58 Wh per imageMeasured2026-10-11link
Qwen-Image 2.1 (edit)ComfyUIMac M4 Pro3 min 52 s / image (768 edit)Measured2026-10-30link
Qwen-Image-2.1-Turbo (official 8 steps)UnspecifiedRTX 509029 s vs 108 s / image (2048²)Measuredlink
Qwen-Image-2.1-TurboUnspecifiedRTX 3060~35 s / image (8 steps)Measuredlink
Z-Image Turbo vs Qwen-Image 2.1UnspecifiedRTX 5090Z-Image ~2× faster (no absolute times); quality: Qwen wins 9:6Measuredlink
Qwen-Image 2.1UnspecifiedRTX 309082–96 s / 2K (public receipt)Relayedlink
Qwen-Image 2.1UnspecifiedRTX 5090 32GB5.93 s; NVFP4 6.7 s vs INT8 7.6 s vs BF16 15.6 s (1024²)Relayedlink
Qwen-Image 2.1UnspecifiedRTX 509013.5 s → 1.9 s / image (vendor claim)Claimedlink
Qwen-Image 2.1UnspecifiedRTX 4060 8GB / 3060 12GB / 5070 Ti 16GB~2 min / ~40 s / ~25 s (1024², 25 steps)Relayedlink
Qwen-Image 2.1SGLang1× DGX Spark (vs 4090)52.6 s / 1024²; SGLang table: Spark 35.4 s, 4090 18.7 sRelayedlink
Qwen-Image 2.1mfluxMac M5 Max~78 s / 1024² (40 steps)Relayedlink

Numbers come from each source. This project did not measure them again. Models and settings differ between rows. Source levels: Official = project or vendor; Measured = the poster ran it (an engine or recipe author must also publish raw data and reproduction steps); Relayed = second-hand; Claimed = prediction or no reproduction details.