Ai makes pictures look like real people.

InsightFaceSwap

Spells used:

A handsome young man, keeping the consistency of action, expression, clothing, shape and appearance of photos, 3d characters from Disney Pixar, super details, blender, soft lighting, ip, blind box, movie edge lighting, romantic scene-v 5. 1-s 750-q 2.

After trying, I come to the conclusion that the current algorithm model can only be based on a single photo (front, high definition, no bangs, no glasses), and the photo angles before and after face changing should be basically the same to get good results. ? I can only say that I have done it, but it is exactly the same. I can't do it now.

Compared with the face randomly generated by midjourney, the similarity has been greatly improved. ? It is believed that with the upgrade of tools, accurate virtual photography with midjourney is just around the corner.

1. In addition to conventional spells, if you want to be more like a person with a cushion, you need to manually add a description of the appearance characteristics of the person, similar to: the next mole on the right eye/round face/small chin/plum cheek/thin lips.

2. Use parameter —iw, with the range of 0.25-2. The closer the value is to 2, the closer it is to a real person, but not a person on the mat. MJ doesn't know much about Asian faces.

3. Use high-definition photos with clear facial features. If the photo itself highlights the characteristics of personal five senses, the picture will be more satisfactory.

4. Keep scrolling until you are satisfied. You can set remix mode in /setting or adjust promt before scrolling again.

5. Use seed parameters for adjustment.

6. Using niji robot mj is more suitable for making two-dimensional drawings.