How to identify mask with real human ?

Currently my model trains the human mask and matches with real human faces. How to avoid this ?

Even if i do video capturing, all his face expressions and angles can be mimiced with human mask if i mistakenly train him as a real person.

Iris unique eye detection also doesnot work. Because a fraud person can keep just his eye holes open in the mask (I dont want to go into this, as i have less knowledge and am not sure if this could also be mimiced with wearing lens)

Speech could also be mimiced by person wearing the mask or a recorded audio.

How to detect human masks from original human faces ?

Currently i feel like only live person finger print could never be mimiced. Let me know if this could also be mimiced. But this is so old and doesnot work good incase of sunlight or body becomes too cold. Have to do again and again till i get recognized.

Sorry for being uncivilized. But only this comes to my mind. Can saliva (easy to generate) / urine (depends on user mood) /(blood (depends on user trust) be used to identify a person in less time ? Hoping no one steals excretions of any human. But this is not computer science anymore. I have no biology background.

Is there anything that captures fraud fastly with only computer science knowledge and be decent and easy to implement at the cost of no inconvience to user ?

I understand what i am asking is beyond this library functionality. But i just want to know your ideas/suggestions. Maybe someone using this libraray had these questions in their mind too.

Thank you,

The issue asks for mask and presentation-attack detection but names no files, tests, or entry points. It provides no concrete implementation scope or criterion for completion, and notes that the request may be beyond this library's functionality.