Chinese female swordmaster (4 different workflows)
Update: v4 of workflow. Image 4:
First pass in Qwen, then did 2nd pass in WAN with 0.5 denoise from qwen image, best results so far
Update: v3 of workflow
Running detail daemon on first pass at 1152x1536 to get a lot more detail even at 10 steps. Euler/Simple.
Then Ultimate SD Upscale Custom Sample node 12 steps, only running at 1.5x to output at 1728x2304 ( Euler/ KARRAS )
Already getting better detail than at 4k output from v2 image.
After that, for 3rd step, I ran it through a FaceDetailer at 1536 max_size to supersample the face for more realism. Better details on eyelids/eyeball/lips. (Euler/Simple)
Then light touch on adaptive sharpen and 0.1 strength LUT after that.
-----------------
Single workflow, low strength film grain and LUT nodes at the end
qwen 6 step, then detail daemon, then upscale sampler.
2nd image is original v1.
3rd image is v2 workflow at 3456x4608, using both Detail Daemon + Ultimate SD in one combined node, to save on using separate nodes.
The focus of this second pic is clarify smaller details in the image output. Such as the hair and clothing embroidery.
I'm uploading it at high res. The DD+upscale process is only using 8 steps, so can see some minor unwanted tiling textures. Acceptable trade off for doing upscaling quicker.
Ultimate SD Upscale (Custom sample) node is using euler, karras 8 steps 0.7 denoise, at 3x upscale, using SwinIR x8 upscale model.





