Hair optimization during video keying

I use the Composure plug-in for video keying. Use a 4K camera to collect a human body model, and the resolution of all materials is 3840*2160. It is found that the hair strand details can be seen in the media_plate, but the details are severely lost in the final synthesized image.
Thanks in advance.

336074-hair0.jpg

336075-hair1.jpg

336076-hair2.jpg