Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when implementing particle effects — GPUParticles2D/3D, ParticleProcessMaterial, emission shapes, subemitters, trails, attractors, collision, and common VFX recipes
.claude/skills/jame581-particles-vfx/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 78% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 146% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 125% | 0% |
All examples target Godot 4.3+ with no deprecated APIs. GDScript is shown first, then C#.
> Related skills: shader-basics for custom particle shaders, 3d-essentials for lighting and environment that affect particles, 2d-essentials for 2D rendering context, tween-animation for code-driven VFX timing, godot-optimization for particle performance tuning.
| Node | Processing | Features | Use For | |---------------------|------------|-----------------------------------------|--------------------------------| | GPUParticles2D | GPU | Full features, high counts, trails | Most 2D effects | | GPUParticles3D | GPU | Full features, attractors, collision | Most 3D effects | | CPUParticles2D | CPU | Simpler, no trails/attractors | Low-end devices, few particles | | CPUParticles3D | CPU | Simpler, no trails/attractors | Low-end devices, few particles |
Rule of thumb: Use GPU particles by default. Switch to CPU particles only for low-end/web targets or when you need CPU-side particle positions (e.g., spawning objects at particle locations).
> You can convert between GPU and CPU particles in the editor: select the node → toolbar → Convert to CPUParticles2D/3D (or vice versa).
GPUParticles2D/3D
├── Process Material (ParticleProcessMaterial) ← physics, emission, color
├── Draw Pass 1 (Mesh) ← what each particle looks like
└── (Optional) Draw Pass 2-4 ← additional meshes| Property | Type | Description | |-------------------|----------|-----------------------------------------------------| | emitting | bool | Start/stop emission | | amount | int | Total particles alive at once | | lifetime | float | Seconds each particle lives | | one_shot | bool | Emit once then stop | | preprocess | float | Simulate this many seconds before first frame | | speed_scale | float | Time multiplier for particle physics | | explosiveness | float | 0.0 = spread over lifetime, 1.0 = all at once | | fixed_fps | int | Lock particle update rate (0 = match render FPS) | | local_coords | bool | Particles move with the node (true) or stay in world (false) | | draw_order | enum | Index, Lifetime, or Reverse Lifetime | | amount_ratio | float | Fraction of particles to emit (0.0–1.0) |
gdscript# Continuous emitter (fire, smoke, ambient dust) $GPUParticles2D.one_shot = false $GPUParticles2D.emitting = true # One-shot burst (explosion, impact splash) $GPUParticles2D.one_shot = true $GPUParticles2D.emitting = false # arm it # Later, trigger: $GPUParticles2D.restart() $GPUParticles2D.emitting = true
csharp// Continuous var particles = GetNode<GpuParticles2D>("GPUParticles2D"); particles.OneShot = false; particles.Emitting = true; // One-shot burst particles.OneShot = true; particles.Emitting = false; // Trigger: particles.Restart(); particles.Emitting = true;
GPUParticles3D gains TRANSFORM_ALIGN_LOCAL_BILLBOARD (= 4): each particle's Z axis faces the camera while preserving a given axis — X or Y, chosen via transform_align_axis. For billboarded particles, transform_align_channel_filter selects which custom channel to read to calculate their angle. ParticleProcessMaterial pairs this with per-axis rotation velocity: enable use_rotation_velocity_3d, then set rotation_velocity_3d_min/max (Vector3, on the particle's local axes) and optionally rotation_velocity_3d_curve (per-axis curve over lifetime).
gdscript# 3D only — billboard toward the camera while keeping the Y axis fixed. # Assumes a ParticleProcessMaterial is assigned (section 1 setup). $GPUParticles3D.transform_align = GPUParticles3D.TRANSFORM_ALIGN_LOCAL_BILLBOARD $GPUParticles3D.transform_align_axis = RenderingServer.PARTICLES_ALIGN_AXIS_Y var mat: ParticleProcessMaterial = $GPUParticles3D.process_material mat.use_rotation_velocity_3d = true mat.rotation_velocity_3d_min = Vector3(-2.0, 0.0, 0.0) mat.rotation_velocity_3d_max = Vector3(2.0, 0.0, 0.0)
csharp// Assumes a ParticleProcessMaterial is assigned (section 1 setup). var particles = GetNode<GpuParticles3D>("GPUParticles3D"); particles.TransformAlign = GpuParticles3D.TransformAlignEnum.LocalBillboard; particles.TransformAlignAxis = RenderingServer.ParticlesTransformAlignAxis.Y; var mat = (ParticleProcessMaterial)particles.ProcessMaterial; mat.UseRotationVelocity3D = true; mat.RotationVelocity3DMin = new Vector3(-2.0f, 0.0f, 0.0f); mat.RotationVelocity3DMax = new Vector3(2.0f, 0.0f, 0.0f);
The material drives per-particle behavior: emission shape (Point / Sphere / Box / Ring / Points / Directed Points), direction + spread + initial velocity, gravity, scale and color over lifetime (via scale_curve / color_ramp), damping, radial/tangential acceleration, and angular velocity.
> See references/process-material-basics.md for the emission-shape table and GDScript + C# snippets for each property group.
Randomize scale and initial orientation per axis instead of uniformly. use_scale_3d enables scale_3d_min/max (Vector3 random scale per particle); use_rotation_3d enables rotation_3d_min/max (Vector3, degrees — works only in 3D).
gdscriptmat.use_scale_3d = true mat.scale_3d_min = Vector3(0.5, 1.0, 0.5) mat.scale_3d_max = Vector3(1.0, 2.0, 1.0) mat.use_rotation_3d = true # 3D only mat.rotation_3d_min = Vector3(0.0, -180.0, 0.0) # degrees mat.rotation_3d_max = Vector3(0.0, 180.0, 0.0)
csharpmat.UseScale3D = true; mat.Scale3DMin = new Vector3(0.5f, 1.0f, 0.5f); mat.Scale3DMax = new Vector3(1.0f, 2.0f, 1.0f); mat.UseRotation3D = true; // 3D only mat.Rotation3DMin = new Vector3(0.0f, -180.0f, 0.0f); // degrees mat.Rotation3DMax = new Vector3(0.0f, 180.0f, 0.0f);
particle_flag_inherit_emitter_scale (default false): if true, particles inherit the scale of the emitter node. Has no effect when local_coords is true, since particles in local space are already affected by the emitter's scale.
gdscriptmat.particle_flag_inherit_emitter_scale = true
csharpmat.ParticleFlagInheritEmitterScale = true;
The recipes most projects need: fire (2D, looped emission with hot-color gradient + scale-down), explosion burst (one-shot, high-amount short-lifetime), dust / footstep puff (one-shot, scale-up + rapid fade).
> See references/vfx-recipes.md for ready-to-use GDScript wiring and recommended ParticleProcessMaterial settings for all three.
Set trail_enabled = true on GPUParticles2D/3D and assign a Mesh (RibbonTrailMesh or TubeTrailMesh). Trails are NOT supported in the Compatibility renderer.
> See references/trails.md for the setup and trail-mesh-type comparison.
A particle can spawn another particle scene at lifecycle events (birth, collision, death, manual). Configure via ParticleProcessMaterial.SubEmitterMode + subemitter property on the parent particles node.
> See references/subemitters.md for trigger modes, scene setup, GDScript and C# (v1.6.0 parity), and limitations.
> ⚠️ Changed in Godot 4.7: Subemitter velocity inheritance was reworked (GH-118062). With sub_emitter_keep_velocity = true (default false), subemitted particles inherit the parent particle's velocity when they spawn. Subemitter effects authored on earlier versions may look different after upgrading — re-check initial velocity and spread on affected systems.
GPUParticlesAttractor*3D (Box / Sphere / Vector Field) pulls particles toward a region. GPUParticlesCollision*3D (Box / Sphere / SDF / HeightField) lets particles bounce off geometry. Both Forward+/Mobile only; no 2D equivalents.
> See references/attractors-and-collision.md for full setup of each attractor and collision type.
Set turbulence_enabled = true on ParticleProcessMaterial and tune turbulence_noise_strength (0.5–2.0 typical), turbulence_noise_scale (lower = larger swirls), turbulence_noise_speed (animate the noise field). Cheap effect for "alive" smoke, fire, dust.
Sprite-sheet animated particles via ParticleProcessMaterial.AnimSpeedMin/Max + CanvasItemMaterial.ParticlesAnimHFrames/VFrames for the sheet layout. Particles cycle through frames over their lifetime.
> See references/flipbook-animation.md for the full setup with GDScript + C# (v1.6.0 parity).
The biggest wins are the obvious ones: keep amount at the minimum that reads well, set fixed_fps = 30 for ambient systems, always set visibility_rect on 2D particles, and expose amount_ratio as a quality slider. Godot 4.7+ adds request_particles_process() for seeking a paused timeline.
Most "broken particles" reports are one of eleven known causes — invisible (no texture / no draw-pass mesh), vanishing (lifetime too short), one-shot not re-firing (needs restart() first), wrong direction (2D Y is inverted), or a base color silently overriding color_ramp.
Full performance table, the 4.7+ timeline-seek API, dynamic quality scaling (GDScript + C#), and the complete symptom/cause/fix table: references/performance-and-pitfalls.md
amount is set to the minimum needed for the visual effectlifetime matches the visual duration — not too short or too longone_shot is enabled for burst effects (explosions, impacts)preprocess is set for always-visible ambient effects (fire, smoke, dust)color_ramp fades alpha to 0 at the end so particles don't vanish abruptlyscale_curve shrinks particles over lifetime for natural fadelocal_coords is set correctly — true for attached effects, false for world-spacequeue_free after lifetime + marginvisibility_rect (2D) is set to prevent particles from being culled prematurelyamount_ratio for player-accessible quality settings| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 16,046 | 8,289 | -48% | 1 | 1 | 0% | 2,815 | 5,013 | +78% | 0 | 0 | — |
case-02 | fail→pass | 33,093 | 5,714 | -83% | 1 | 1 | 0% | 5,997 | 4,366 | -27% | 0 | 0 | — |
case-03 | pass→pass | 20,959 | 14,857 | -29% | 1 | 1 | 0% | 3,454 | 5,739 | +66% | 0 | 0 | — |
case-04 | pass→pass | 20,177 | 17,980 | -11% | 1 | 1 | 0% | 3,597 | 6,703 | +86% | 0 | 0 | — |
case-05 | pass→pass | 10,487 | 7,809 | -26% | 1 | 1 | 0% | 1,867 | 4,725 | +153% | 0 | 0 | — |
case-06 | pass→pass | 15,664 | 14,262 | -9% | 1 | 1 | 0% | 2,421 | 5,522 | +128% | 0 | 0 | — |
case-07 | fail→pass | 10,705 | 5,878 | -45% | 1 | 1 | 0% | 1,711 | 4,215 | +146% | 0 | 0 | — |
case-08 | fail→pass | 24,312 | 3,990 | -84% | 1 | 1 | 0% | 4,082 | 4,065 | -0% | 0 | 0 | — |
case-09 | pass→pass | 8,763 | 4,332 | -51% | 1 | 1 | 0% | 1,411 | 4,158 | +195% | 0 | 0 | — |
case-10 | pass→pass | 6,186 | 5,240 | -15% | 1 | 1 | 0% | 965 | 4,204 | +336% | 0 | 0 | — |
case-11 | pass→pass | 8,290 | 4,879 | -41% | 1 | 1 | 0% | 1,479 | 4,157 | +181% | 0 | 0 | — |
case-12 | fail→pass | 10,024 | 3,551 | -65% | 1 | 1 | 0% | 1,728 | 3,887 | +125% | 0 | 0 | — |
case-13 | fail→fail | 10,837 | 6,573 | -39% | 1 | 1 | 0% | 1,825 | 4,424 | +142% | 0 | 0 | — |
case-14 | fail→pass | 24,423 | 4,772 | -80% | 1 | 1 | 0% | 4,326 | 4,169 | -4% | 0 | 0 | — |
case-15 | pass→pass | 4,340 | 4,349 | +0% | 1 | 1 | 0% | 733 | 4,093 | +458% | 0 | 0 | — |
case-16 | pass→pass | 10,656 | 9,561 | -10% | 1 | 1 | 0% | 1,703 | 4,961 | +191% | 0 | 0 | — |
case-17 | pass→pass | 13,496 | 8,094 | -40% | 1 | 1 | 0% | 2,218 | 4,736 | +114% | 0 | 0 | — |
case-18 | pass→pass | 15,777 | 15,947 | +1% | 1 | 1 | 0% | 2,617 | 5,999 | +129% | 0 | 0 | — |
case-19 | pass→pass | 5,164 | 7,281 | +41% | 1 | 1 | 0% | 827 | 4,794 | +480% | 0 | 0 | — |
case-20 | pass→pass | 5,799 | 3,699 | -36% | 1 | 1 | 0% | 919 | 3,964 | +331% | 0 | 0 | — |
case-21 | pass→pass | 8,404 | 5,106 | -39% | 1 | 1 | 0% | 1,314 | 4,215 | +221% | 0 | 0 | — |
case-22 | fail→pass | 30,997 | 4,022 | -87% | 1 | 1 | 0% | 5,585 | 4,034 | -28% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.