NFS MW and Carbon Camera Mods

• UG2 style camera for MW & CB
• Cinematic camera for CB

In Need for Speed: Most Wanted (2005) the camera system is multi-stage, and depending on where you hook, it looks like two passes per frame.

Here’s the precise breakdown.

This happens during the simulation update:

• Camera director logic runs (follow cam, drift cam, cinematic blends)

• Car state is fresh (physics just updated)

• Look-at matrices are built for intent

• This is where:

  • yaw rate
  • velocity direction
  • roll intent should be sampled

Your hook trigger:

hkCubicUpdate → gApplyUG2Flag = 1

is correctly marking this pass.

Later in the same frame (sometimes multiple times):

• Camera matrix is rebuilt or touched again

• Used for:

  • motion blur
  • mirrors
  • reflections
  • post effects
  • UI overlays

• Physics is not updated

• Director may not run

This pass:

• must NOT drive camera dynamics
• must NOT integrate roll/yaw again
• must only consume already-computed camera state

Because:

• CreateLookAt (or equivalent) can be called more than once per frame
• But only one of those calls is authoritative

So if you:

• integrate roll/yaw on every call → jitter, edgy feel
• integrate only on flagged call → smooth (UG2 behavior)

You already discovered this empirically with gApplyUG2Flag.

Think of MW camera like this:

[ Sim Update ]
   └─ Director computes camera intent
      └─ (you sample & integrate dynamics here)

[ Render Pass ]
   ├─ Rebuild view matrix
   ├─ Rebuild view matrix again (mirrors / blur)
   └─ Use same camera state

So:

• State update = once per frame
• Matrix application = many times per frame

UG2 explicitly separates:

• camera state integration (once per frame)
• view matrix emission (many times)

MW does the same implicitly — you just have to respect it when hooking.

Your current structure is now correct because:

bool applyRoll = (InterlockedExchange(&gApplyUG2Flag, 0) != 0);

// state update → always
// roll application → only if applyRoll

That is exactly what MW expects.

If you accidentally:

• integrate yaw/roll on non-authoritative passes
• or reset state when applyRoll == false

you get:

• micro spikes
• jitter
• snapping on transitions

90% fixed of this by:

• decoupling state update vs application
• adding hard-cut detection
• smoothing rollTarget

• MW does not do two full camera updates per frame
• It does emit the camera matrix multiple times
• Only one pass per frame should drive dynamics
• Your gApplyUG2Flag approach is correct
• Remaining “edginess” is signal shaping, not engine behavior