How to avoid the offset between two freelook virtual cameras in a tps camera system?

What do you mean when you say "screen x offset with a state driven camera"?
How do you implement the screen X offset?

The only way to get what you're looking for is to ensure that when you switch between Aim and Normal, the vcam does not rotate at all, and does not move except along its Z axis. This can be done with CM, but you have to take considerable care with the Aim settings and the orbits.

 

You need to be careful with your settings.
First, and most importantly, make sure that both FreeLooks have InheritPosition checked:



This forces the incoming FreeLook to be warped to the correct place on the orbit to match the outgoing.

Next, you have to tune the screen position. Because you're changing the focal length and because the target is off-center, if you want the center of the screen to remain fixed you can't have the target at a constant screen point across different FOVs - it's physically impossible unless the FOV remains fixed. Cinemachine puts the LookAt target at the point you specify. It doesn't care about screen center.

In your example, I tried setting the screen X to 0.35 for the normal cam, and 0.25 for the Aim cam. That came pretty close. You'll have to do similar tuning for the Y position.

A better and less finicky approach to solving this problem is to put all targets at screen center, then add a CM extension to shift the whole camera at the end of the CM pipeline, after all the calculations have been done.

Here is an extension that does that. You can add it to the SDC, and it will affect all the child FreeLooks.

Code (CSharp):

1. using UnityEngine;
2. using Cinemachine;
3.  
4. /// <summary>
5. /// An add-on module for Cinemachine Virtual Camera that adds a final offset to the camera
6. /// </summary>
7. [AddComponentMenu("")] // Hide in menu
8. public class CinemachineCameraOffset : CinemachineExtension
9. {
10.     [Tooltip("Offset the camera's position by this much (camera space)")]
11.     public Vector3 m_Offset = Vector3.zero;
12.  
13.     protected override void PostPipelineStageCallback(
14.         CinemachineVirtualCameraBase vcam,
15.         CinemachineCore.Stage stage, ref CameraState state, float deltaTime)
16.     {
17.         // Apply after the camera has been aimed.
18.         // To apply offset before the camera aims, change this to Body
19.         if (stage == CinemachineCore.Stage.Aim)
20.         {
21.             Vector3 offset = state.FinalOrientation * m_Offset;
22.             state.ReferenceLookAt += offset;
23.             state.PositionCorrection += offset;
24.         }
25.     }
26. }

Finally, you'll get best results if you add an empty game object inside your character, to serve as a LookAt and Follow target. Place it in the center of the head, and set all the LookAt offsets in the vcam to 0. Adjust your orbits to accommodate the new target pos. Having the target be actually at the center of the orbits (instead of using orbits that are offset vertically as you have) will maximize CM's ability to blend seamlessly between FreeLook targeting that character.

 

In your project, you did 2 things to make the Offset script not work properly:

1. You changed the CameraOffset script to apply the offset after the Body stage. To get the correct behaviour, you have to leave it the way it was (i.e. apply the offset after the Aim stage)
2. You changed the FreeLooks to SimpleFollow binding mode. An X offset won't work in SimpleFollow. You have to leave the binding mode on WorldSpace.