Fix Godot MultiMesh Frustum Culling Disabled Instances

MultiMesh is Godot’s fast path for rendering thousands of identical meshes in one draw call. The tradeoff is that it is one draw call — one renderer, one AABB, one visibility test. If you put ten thousand trees into a single MultiMesh and the camera can see even one of them, the GPU renders the full ten thousand. Performance does not scale with screen coverage, and that surprises most people the first time they profile a forest.

Confirm what Godot sees as the bounding box

Open the scene, select the MultiMeshInstance3D, and look at the Custom AABB property in the Inspector. If it is the default zero-size box, Godot is deriving the AABB from the instance transforms at load time. For a buffer you populate from code, that derivation may run before the transforms are written, leaving you with a collapsed AABB centered on the origin. Symptom: the node is culled whenever the camera looks away from the origin, even though instances are spread across the level.

Write a correct AABB explicitly after you fill the transforms:

func populate() -> void:
    var mm := multimesh
    mm.instance_count = instance_count
    var aabb := AABB(Vector3.ZERO, Vector3.ZERO)
    for i in instance_count:
        var t := Transform3D(Basis.IDENTITY, positions[i])
        mm.set_instance_transform(i, t)
        aabb = aabb.expand(positions[i])
    custom_aabb = aabb.grow(mesh_radius)

The grow accounts for the radius of each mesh so the edge of the frustum does not clip instances whose center is just outside the box.

Partition into cells

Once the AABB is correct the node is cullable, but it is still all-or-nothing. A forest of 10,000 trees in one MultiMeshInstance3D renders all 10,000 if even one is visible. For real savings, split the world into a grid:

const CELL_SIZE := 32.0

func build_cells(trees: Array) -> void:
    var buckets := {}
    for t in trees:
        var key := Vector2i(floor(t.x / CELL_SIZE), floor(t.z / CELL_SIZE))
        if not buckets.has(key): buckets[key] = []
        buckets[key].append(t)
    for key in buckets:
        _spawn_multimesh(buckets[key])

A 32m cell size is a reasonable starting point for dense vegetation. Each cell becomes a separate MultiMeshInstance3D with its own AABB, and Godot culls each one independently against the frustum. Frame time scales with the number of visible cells instead of the total instance count.

Add visibility range for distant chunks

Frustum culling only hides what is off-screen. You still pay for cells that are on-screen but beyond a reasonable draw distance. Use the GeometryInstance3D visibility range fields on each MultiMeshInstance3D:

visibility_range_end = 120.0
visibility_range_end_margin = 20.0
visibility_range_fade_mode = GeometryInstance3D.VISIBILITY_RANGE_FADE_SELF

Godot will smoothly fade cells out as the camera backs away. Combined with a lower-LOD billboard MultiMesh for long range, you keep visual density without paying for full-resolution meshes across the entire world.

Per-instance culling is not built in

A frequent ask is to have Godot skip individual instances inside a MultiMesh based on the frustum. The engine does not do this in the standard pipeline because the whole point of MultiMesh is to avoid per-instance work on the CPU. If you need per-instance culling — for example, to disable grass instances behind hills — you have to implement it manually with a compute shader that writes indirect draw arguments, which is an expert-level path.

Verify in the profiler

Open the Monitor tab and watch Draw Calls (3D) and Vertices Drawn. Before partitioning you will see vertices equal to instances times mesh verts, regardless of camera angle. After partitioning, turning the camera should drop the vertex count immediately as cells leave the frustum. If it does not, your AABBs are still wrong.

Understanding the issue

This bug class falls into a pattern that's worth understanding beyond the specific case. In Godot Engine, the underlying behavior is shaped by how the engine layers its abstractions - the public API you call, the runtime systems that respond, and the platform-specific implementations underneath. A bug at any layer can produce symptoms that look like they originate at a different layer. Triaging effectively means recognizing which layer the symptom belongs to, even when the gameplay code is what's visible.

The specific bug described above is the kind that surfaces during integration rather than unit testing. It depends on a combination of factors: the asset configuration, the runtime state, the platform's specific behavior. In isolation, each piece looks correct; in combination, the bug emerges. This is why thorough integration testing - playing the actual game in realistic conditions - catches things that automated tests miss.

Why this happens

Bugs of this class are particularly easy to ship past internal QA because they often depend on specific runtime conditions - hardware combinations, network states, or asset configurations that QA didn't reproduce. Players hit them in the wild, file reports that are hard to repro, and the bug accumulates negative reviews while engineering tries to recreate the failure mode.

At the engine level, the behavior comes from a deliberate design decision in Godot. The engine team chose a particular trade-off - usually performance versus convenience, or generality versus specificity - and that trade-off has consequences when you push against it. Understanding the trade-off is what turns 'this bug is mysterious' into 'this bug is the expected consequence of this design'.

Verifying the fix

After applying the fix, the verification step has three parts: confirm the original repro is resolved, confirm no obvious regressions in adjacent functionality, and (for shipping titles) deploy to a small player cohort first and watch the crash and report rates. Each step catches something the others miss.

Reproducibility is the prerequisite for verification. If you can't reliably reproduce the bug pre-fix, you can't reliably verify it post-fix. Spend time getting a clean reproduction before you write any fix code. The fix is fast once you understand the reproduction; the reproduction is the slow part.

Variations to watch for

Related bug classes often share the same root cause. If you find yourself fixing this issue, look for cousins: similar symptoms in adjacent systems, the same data flow but a different value, or the same fix pattern in another module. The catalog of 'we've seen this before' becomes valuable institutional knowledge.

Adjacent bugs often share a root cause. After fixing the case you've found, spend an hour searching the codebase for similar patterns. What's the same call with different arguments? The same data flow with a different entity type? The same lifecycle issue in a sibling system? Each match is a candidate for the same fix, or a related fix that prevents future bugs of the same class.

In production

For shipping titles with a long support window, watch for this issue resurfacing after dependency updates. Engine upgrades, driver updates, OS releases - each one can resurface a bug class you thought you'd fixed because the underlying behavior changed slightly. Regression tests catch the obvious ones; player reports catch the rest.

When triaging a similar issue in production, prioritize gathering data over hypothesizing causes. A player report describes a symptom; what you need is a build SHA, a session timestamp, and ideally a screen recording or session replay. With those, the bug becomes tractable. Without them, you're guessing at hypothetical reproductions that may not match what the player actually hit.

Performance considerations

If this issue manifests under high load (many actors, many particles, many network connections), profile the post-fix code path with realistic counts. The original cost was a bug; the new cost is real work, and real work has a budget.

Diagnostic approach

Before applying any fix, gather enough context to be confident you're addressing the actual cause and not a similar-looking symptom. The cheapest diagnostic step is reproducing the bug deterministically - if you can't get the same failure twice in a row, your fix attempts will be hard to evaluate. Lock down the reproduction first.

For Godot-specific diagnostics, the editor's profiler is the canonical starting point. Capture a representative frame with the symptom present; compare against a frame without the symptom; the diff often points directly at the cause. If the symptom is non-deterministic, capture multiple frames and look for the pattern - the cause is usually a state transition or a specific input value rather than a continuous effect.

Tooling and ecosystem

The tooling around this bug class matters as much as the fix itself. Good logging, accessible profilers, and clear error messages turn 30-minute investigations into 5-minute ones. If your project doesn't have visibility into this code path, the first fix should add the visibility - the second fix uses it.

Within Godot, the relevant diagnostic surfaces include the standard frame debugger, memory profiler, and engine-specific debug overlays. Each one shows a different facet of what's happening. The frame debugger reveals draw call ordering and state transitions; the memory profiler shows allocation patterns; the debug overlay reveals per-system state. Bugs that resist one tool usually surrender to another - the trick is knowing which tool to reach for first.

Edge cases and pitfalls

Boundary conditions deserve specific testing attention. What happens when the input is zero, maximum, negative, or NaN? What happens at the start of a session vs hours in? What happens at the boundary between two systems handling the same data? These are where bugs hide and where regression tests are most valuable.

When writing a regression test for this fix, focus on the boundary conditions that surfaced the original bug. Tests that exercise the happy path catch obvious regressions; tests that exercise the boundary catch the subtler regressions that look like new bugs but are really the original returning. The latter are the tests that earn their keep over the long life of the project.

Team communication

When this bug class affects multiple teams (often the case for cross-system issues), early communication prevents duplicate work. The team that owns the symptom may not own the cause. A 15-minute conversation at the start of triage often saves hours of independent investigation.

If this fix touches a system several engineers work in, a short writeup in the team's engineering channel helps. Not a full design doc - a paragraph explaining what was wrong, what's fixed, and what to watch for. Future engineers encountering similar symptoms will search for the fix; making it findable is a small investment that pays back later.

“A MultiMesh is one cullable object to Godot. Want finer culling? Give Godot more objects by splitting the buffer.”

Related Issues

For related rendering performance work, see Fix Godot 3D spatial audio not attenuating, and for viewport-related rendering gotchas, Fix Godot viewport stretch mode black bars unexpected.

Tip: visualize the AABB in the editor — if it looks like a thin slice instead of the full distribution, it is stale.