QWEN LORA LOADER:  https://github.com/Cx330v/Qwen-Image-Toolkit

QWEN REALISM LORAS ARE OUT:  and just 15 steps

cfg: 2.0

sampler: res_2m

scheduler: bong_tangent

shift: 7:00

resolution: 1328x928

Prompt executed in 153.95 seconds on an RTX 560 ti 16GB and 64GB system Ram.

You may halve the execution time by lowering CFG to 1.0 but with small details loss.

MODELS:

QWEN-IMAGE FP8 https://huggingface.co/.../qwen_image_fp8_e4m3fn.safetensors

TEXT ENCODER FP16 or FPH https://huggingface.co/.../qwen_2.5_vl_7b_fp8_scaled...

VAE https://huggingface.co/.../vae/qwen_image_vae.safetensors

REALISM LORA:  https://huggingface.co/flymy-ai/qwen-image-lora/blob/main/pytorch_lora_weights.safetensors download and rename.