Perhaps it would help your understanding if you consider that VLANs generally create virtual switches.

If one switch has two VLANs defined, it logically operates as two physical switches. Traffic within the switch from one VLAN is not visible to the other VLAN, just as traffic between two physical switches isn't visible to each other.

However, this isolation is broken, once we start interconnecting switches. Two non-VLAN switches, interconnected will be the part of the same LAN segment. We could likewise interconnect, on the same switch, a port from one VLAN to a port of the second VLAN and they then would also be joined.

If you had two switches, each with two VLANs, again, the VLANs are isolated on each switch until we start to interconnect. We could interconnect switch1's one VLAN with one VLAN on switch2 and interconnect switch1's second VLAN with the second VLAN on switch2. Now we have the equivalent of 4 physical switches, divided into connected pairs.

If we interconnect the two VLAN on either physical switch (just one interconnect), we then have the equivalent of 4 physically connected switches. If the interconnection was on switch2, traffic that wanted to go from one switch1 VLAN to the other would first pass to switch2, jump the VLANs there, and come back to switch1.

Since we normally use VLANs to logically contain a segment, interconnecting VLANs, as described above, defeats their purpose. I.e. if we wanted traffic to freely mingle between two VLANs, why define them. Instead, we normally join VLANs with a router, just as a router could physically join separate non-VLAN switches.

PS:

One feature also used with VLANs is "trunk" connections. Instead of having a physical connection to support each VLAN crossing between devices, frames are tagged with a VLAN ID. This creates virtual links.