1.Face first, then clothing – Replace the face first, then change the outfit.

2.Models used – Face swapping is done using ACELora's Portrait, while clothing replacement uses Subject.

3.Flux-Turbo for efficiency – Uses Flux-Turbo with 8 steps to reduce generation time and stabilize the image. Each process runs for 12 steps.

4.LayerStyle for masking – LayerStyle is used to cut out face and clothing masks. Beginners often encounter installation issues with this node, but it performs better than SAM for mask segmentation. Learn to overcome these challenges on your own.

5.Face swapping without Redux – Only prompt-based face swapping is used to set facial expressions. For example, if the prompt includes "Smiling," the character will smile; if "Sleeping" is used, the character will close their eyes. If Redux were used, the expression would strictly match the input image, ignoring the prompt.

6.Clothing replacement with Redux – Redux is used to maintain clothing integrity.

7.ICMask for segmentation – The ICMask node is used to separate input and output images, ensuring a seamless transformation.

8.Preserving details – In the final output, almost all text and images on the Wikipedia clothing are retained.

9.A Example of Loop in ACELora.

Tools Works On TensorArt

https://tensor.art/template/839749506311406568