So based on data from a quick Google search, what really makes facial recognition secure or not is how the device reads your face.
In instances like mustang7757's with the OP7P and Note 10+, they are considered not secure because they use the front-facing camera to identify your face, which is a 2d image. Like a number of you have said, this means someone can easily spoof it with a picture of you.
On the other hand, what Apple's FaceID does, and what I expect the Pixel 4 will be doing is using 3D IR mapping to identify you. This uses thousands of little IR dots on your face to map a 3D image to figure out if it's you. A picture can't fake this. Additionally, what will make it harder for a twin or relative to fool it will be the algorithm that is used. The device will need to learn the different variations of your face with age, orientation (so you don't have to be straight on every time), etc, and use it to know what slight variations are yours and which aren't (for example, you wouldn't magically grow a mole on your cheek over the course of an hour, so that mustn't be you).
The last component will be how the device uses your face as a key. For it to be secure, it would need to be entirely local to the device, and never be uploaded to the cloud. But Google and Apple have been doing this with secure chips and fingerprints for years, so I'm assuming that will be held constant when switching to face ID.
TL;DR: The hardware that the device uses, and what algorithm will make the difference between secure and not, so some devices are quite secure, and others aren't. Read above to know the difference.