Style + Composition | Supports Online Generation Version 2.0

Usage Methods and Precautions:

This workflow uses the XL model. The weight of the IPAdapter style synthesis SDXL node can be around 0.4-1.

If you are not performing human style transfer, you can ignore the FaceID node (👇 right-click on the top-left corner to ignore this group of nodes), maximize the weight of the IPAdapter style synthesis SDXL, and reduce the weight of the ControlNet application (advanced) to give AI more room for creativity (adjust the start and end times as needed).

If you are performing product style transfer, set the weight of the ControlNet application (advanced) to 1 to control the product, with the start time at 0 and the end time at 1, and the weight of the IPAdapter style synthesis SDXL between 0.4-1.

If you are performing human migration, you will notice that the face is different from the original image after migration. If you care about these differences, save the image after facial repair and then proceed with face swapping.

If you do not want to use the 1.5 model, you can change it to another model, but ensure that the subsequent models are also changed to match, otherwise, it may cause an error and fail to run.

Remember to change the seed of the K-Sampler each time you adjust the parameters, as there may occasionally be a bug. You can change the seed number to any arbitrary value such as 666, 777, 888, or 888888, as long as it is different from the original.

If you have any questions about the above, you can ask in the comments section, and I will reply to everyone as soon as I see it.

Finally, I hope everyone has a good time.

Note: ControlNet should not be fully activated, as excessive use can cause image distortion.