reze-engine 0.1.12 → 0.1.13

package/dist/engine.js

@@ -81,239 +81,120 @@ export class Engine {
         });
         const shaderModule = this.device.createShaderModule({
             label: "model shaders",
-            code: /* wgsl */ `
-        struct CameraUniforms {
-          view: mat4x4f,
-          projection: mat4x4f,
-          viewPos: vec3f,
-          _padding: f32,
-        };
-
-        struct Light {
-          direction: vec3f,
-          _padding1: f32,
-          color: vec3f,
-          intensity: f32,
-        };
-
-        struct LightUniforms {
-          ambient: f32,
-          lightCount: f32,
-          _padding1: f32,
-          _padding2: f32,
-          lights: array<Light, 4>,
-        };
-
-        struct MaterialUniforms {
-          alpha: f32,
-          rimIntensity: f32,
-          rimPower: f32,
-          _padding1: f32,
-          rimColor: vec3f,
-          _padding2: f32,
-        };
-
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-          @location(0) normal: vec3f,
-          @location(1) uv: vec2f,
-          @location(2) worldPos: vec3f,
-        };
-
-        @group(0) @binding(0) var<uniform> camera: CameraUniforms;
-        @group(0) @binding(1) var<uniform> light: LightUniforms;
-        @group(0) @binding(2) var diffuseTexture: texture_2d<f32>;
-        @group(0) @binding(3) var diffuseSampler: sampler;
-        @group(0) @binding(4) var<storage, read> skinMats: array<mat4x4f>;
-        @group(0) @binding(5) var toonTexture: texture_2d<f32>;
-        @group(0) @binding(6) var toonSampler: sampler;
-        @group(0) @binding(7) var<uniform> material: MaterialUniforms;
-
-        @vertex fn vs(
-          @location(0) position: vec3f,
-          @location(1) normal: vec3f,
-          @location(2) uv: vec2f,
-          @location(3) joints0: vec4<u32>,
-          @location(4) weights0: vec4<f32>
-        ) -> VertexOutput {
-          var output: VertexOutput;
-          let pos4 = vec4f(position, 1.0);
-          
-          // Normalize weights to ensure they sum to 1.0 (handles floating-point precision issues)
-          let weightSum = weights0.x + weights0.y + weights0.z + weights0.w;
-          var normalizedWeights: vec4f;
-          if (weightSum > 0.0001) {
-            normalizedWeights = weights0 / weightSum;
-          } else {
-            normalizedWeights = vec4f(1.0, 0.0, 0.0, 0.0);
-          }
-          
-          var skinnedPos = vec4f(0.0, 0.0, 0.0, 0.0);
-          var skinnedNrm = vec3f(0.0, 0.0, 0.0);
-          for (var i = 0u; i < 4u; i++) {
-            let j = joints0[i];
-            let w = normalizedWeights[i];
-            let m = skinMats[j];
-            skinnedPos += (m * pos4) * w;
-            let r3 = mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz);
-            skinnedNrm += (r3 * normal) * w;
-          }
-          let worldPos = skinnedPos.xyz;
-          output.position = camera.projection * camera.view * vec4f(worldPos, 1.0);
-          output.normal = normalize(skinnedNrm);
-          output.uv = uv;
-          output.worldPos = worldPos;
-          return output;
-        }
-
-        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
-          let n = normalize(input.normal);
-          let albedo = textureSample(diffuseTexture, diffuseSampler, input.uv).rgb;
-
-          var lightAccum = vec3f(light.ambient);
-          let numLights = u32(light.lightCount);
-          for (var i = 0u; i < numLights; i++) {
-            let l = -light.lights[i].direction;
-            let nDotL = max(dot(n, l), 0.0);
-            let toonUV = vec2f(nDotL, 0.5);
-            let toonFactor = textureSample(toonTexture, toonSampler, toonUV).rgb;
-            let radiance = light.lights[i].color * light.lights[i].intensity;
-            lightAccum += toonFactor * radiance * nDotL;
-          }
-          
-          // Rim light calculation
-          let viewDir = normalize(camera.viewPos - input.worldPos);
-          var rimFactor = 1.0 - max(dot(n, viewDir), 0.0);
-          rimFactor = pow(rimFactor, material.rimPower);
-          let rimLight = material.rimColor * material.rimIntensity * rimFactor;
-          
-          let color = albedo * lightAccum + rimLight;
-          let finalAlpha = material.alpha;
-          if (finalAlpha < 0.001) {
-            discard;
-          }
-          
-          return vec4f(clamp(color, vec3f(0.0), vec3f(1.0)), finalAlpha);
-        }
-      `,
-        });
-        // Unified hair shader that can handle both over-eyes and over-non-eyes cases
-        // Uses material.alpha multiplier to control opacity (0.5 for over-eyes, 1.0 for over-non-eyes)
-        const hairShaderModule = this.device.createShaderModule({
-            label: "unified hair shaders",
-            code: /* wgsl */ `
-        struct CameraUniforms {
-          view: mat4x4f,
-          projection: mat4x4f,
-          viewPos: vec3f,
-          _padding: f32,
-        };
-
-        struct Light {
-          direction: vec3f,
-          _padding1: f32,
-          color: vec3f,
-          intensity: f32,
-        };
-
-        struct LightUniforms {
-          ambient: f32,
-          lightCount: f32,
-          _padding1: f32,
-          _padding2: f32,
-          lights: array<Light, 4>,
-        };
-
-        struct MaterialUniforms {
-          alpha: f32,
-          alphaMultiplier: f32, // New: multiplier for alpha (0.5 for over-eyes, 1.0 for over-non-eyes)
-          rimIntensity: f32,
-          rimPower: f32,
-          rimColor: vec3f,
-          _padding1: f32,
-        };
-
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-          @location(0) normal: vec3f,
-          @location(1) uv: vec2f,
-          @location(2) worldPos: vec3f,
-        };
-
-        @group(0) @binding(0) var<uniform> camera: CameraUniforms;
-        @group(0) @binding(1) var<uniform> light: LightUniforms;
-        @group(0) @binding(2) var diffuseTexture: texture_2d<f32>;
-        @group(0) @binding(3) var diffuseSampler: sampler;
-        @group(0) @binding(4) var<storage, read> skinMats: array<mat4x4f>;
-        @group(0) @binding(5) var toonTexture: texture_2d<f32>;
-        @group(0) @binding(6) var toonSampler: sampler;
-        @group(0) @binding(7) var<uniform> material: MaterialUniforms;
-
-        @vertex fn vs(
-          @location(0) position: vec3f,
-          @location(1) normal: vec3f,
-          @location(2) uv: vec2f,
-          @location(3) joints0: vec4<u32>,
-          @location(4) weights0: vec4<f32>
-        ) -> VertexOutput {
-          var output: VertexOutput;
-          let pos4 = vec4f(position, 1.0);
-          
-          let weightSum = weights0.x + weights0.y + weights0.z + weights0.w;
-          var normalizedWeights: vec4f;
-          if (weightSum > 0.0001) {
-            normalizedWeights = weights0 / weightSum;
-          } else {
-            normalizedWeights = vec4f(1.0, 0.0, 0.0, 0.0);
-          }
-          
-          var skinnedPos = vec4f(0.0, 0.0, 0.0, 0.0);
-          var skinnedNrm = vec3f(0.0, 0.0, 0.0);
-          for (var i = 0u; i < 4u; i++) {
-            let j = joints0[i];
-            let w = normalizedWeights[i];
-            let m = skinMats[j];
-            skinnedPos += (m * pos4) * w;
-            let r3 = mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz);
-            skinnedNrm += (r3 * normal) * w;
-          }
-          let worldPos = skinnedPos.xyz;
-          output.position = camera.projection * camera.view * vec4f(worldPos, 1.0);
-          output.normal = normalize(skinnedNrm);
-          output.uv = uv;
-          output.worldPos = worldPos;
-          return output;
-        }
-
-        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
-          let n = normalize(input.normal);
-          let albedo = textureSample(diffuseTexture, diffuseSampler, input.uv).rgb;
-
-          var lightAccum = vec3f(light.ambient);
-          let numLights = u32(light.lightCount);
-          for (var i = 0u; i < numLights; i++) {
-            let l = -light.lights[i].direction;
-            let nDotL = max(dot(n, l), 0.0);
-            let toonUV = vec2f(nDotL, 0.5);
-            let toonFactor = textureSample(toonTexture, toonSampler, toonUV).rgb;
-            let radiance = light.lights[i].color * light.lights[i].intensity;
-            lightAccum += toonFactor * radiance * nDotL;
-          }
-          
-          // Rim light calculation
-          let viewDir = normalize(camera.viewPos - input.worldPos);
-          var rimFactor = 1.0 - max(dot(n, viewDir), 0.0);
-          rimFactor = pow(rimFactor, material.rimPower);
-          let rimLight = material.rimColor * material.rimIntensity * rimFactor;
-          
-          let color = albedo * lightAccum + rimLight;
-          let finalAlpha = material.alpha * material.alphaMultiplier;
-          if (finalAlpha < 0.001) {
-            discard;
-          }
-          
-          return vec4f(clamp(color, vec3f(0.0), vec3f(1.0)), finalAlpha);
-        }
+            code: /* wgsl */ `
+        struct CameraUniforms {
+          view: mat4x4f,
+          projection: mat4x4f,
+          viewPos: vec3f,
+          _padding: f32,
+        };
+
+        struct Light {
+          direction: vec3f,
+          _padding1: f32,
+          color: vec3f,
+          intensity: f32,
+        };
+
+        struct LightUniforms {
+          ambient: f32,
+          lightCount: f32,
+          _padding1: f32,
+          _padding2: f32,
+          lights: array<Light, 4>,
+        };
+
+        struct MaterialUniforms {
+          alpha: f32,
+          alphaMultiplier: f32,
+          rimIntensity: f32,
+          rimPower: f32,
+          rimColor: vec3f,
+          _padding1: f32,
+        };
+
+        struct VertexOutput {
+          @builtin(position) position: vec4f,
+          @location(0) normal: vec3f,
+          @location(1) uv: vec2f,
+          @location(2) worldPos: vec3f,
+        };
+
+        @group(0) @binding(0) var<uniform> camera: CameraUniforms;
+        @group(0) @binding(1) var<uniform> light: LightUniforms;
+        @group(0) @binding(2) var diffuseTexture: texture_2d<f32>;
+        @group(0) @binding(3) var diffuseSampler: sampler;
+        @group(0) @binding(4) var<storage, read> skinMats: array<mat4x4f>;
+        @group(0) @binding(5) var toonTexture: texture_2d<f32>;
+        @group(0) @binding(6) var toonSampler: sampler;
+        @group(0) @binding(7) var<uniform> material: MaterialUniforms;
+
+        @vertex fn vs(
+          @location(0) position: vec3f,
+          @location(1) normal: vec3f,
+          @location(2) uv: vec2f,
+          @location(3) joints0: vec4<u32>,
+          @location(4) weights0: vec4<f32>
+        ) -> VertexOutput {
+          var output: VertexOutput;
+          let pos4 = vec4f(position, 1.0);
+          
+          // Normalize weights to ensure they sum to 1.0 (handles floating-point precision issues)
+          let weightSum = weights0.x + weights0.y + weights0.z + weights0.w;
+          var normalizedWeights: vec4f;
+          if (weightSum > 0.0001) {
+            normalizedWeights = weights0 / weightSum;
+          } else {
+            normalizedWeights = vec4f(1.0, 0.0, 0.0, 0.0);
+          }
+          
+          var skinnedPos = vec4f(0.0, 0.0, 0.0, 0.0);
+          var skinnedNrm = vec3f(0.0, 0.0, 0.0);
+          for (var i = 0u; i < 4u; i++) {
+            let j = joints0[i];
+            let w = normalizedWeights[i];
+            let m = skinMats[j];
+            skinnedPos += (m * pos4) * w;
+            let r3 = mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz);
+            skinnedNrm += (r3 * normal) * w;
+          }
+          let worldPos = skinnedPos.xyz;
+          output.position = camera.projection * camera.view * vec4f(worldPos, 1.0);
+          output.normal = normalize(skinnedNrm);
+          output.uv = uv;
+          output.worldPos = worldPos;
+          return output;
+        }
+
+        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+          let n = normalize(input.normal);
+          let albedo = textureSample(diffuseTexture, diffuseSampler, input.uv).rgb;
+
+          var lightAccum = vec3f(light.ambient);
+          let numLights = u32(light.lightCount);
+          for (var i = 0u; i < numLights; i++) {
+            let l = -light.lights[i].direction;
+            let nDotL = max(dot(n, l), 0.0);
+            let toonUV = vec2f(nDotL, 0.5);
+            let toonFactor = textureSample(toonTexture, toonSampler, toonUV).rgb;
+            let radiance = light.lights[i].color * light.lights[i].intensity;
+            lightAccum += toonFactor * radiance * nDotL;
+          }
+          
+          // Rim light calculation
+          let viewDir = normalize(camera.viewPos - input.worldPos);
+          var rimFactor = 1.0 - max(dot(n, viewDir), 0.0);
+          rimFactor = pow(rimFactor, material.rimPower);
+          let rimLight = material.rimColor * material.rimIntensity * rimFactor;
+          
+          let color = albedo * lightAccum + rimLight;
+          let finalAlpha = material.alpha * material.alphaMultiplier;
+          if (finalAlpha < 0.001) {
+            discard;
+          }
+          
+          return vec4f(clamp(color, vec3f(0.0), vec3f(1.0)), finalAlpha);
+        }
       `,
         });
         // Create explicit bind group layout for all pipelines using the main shader
@@ -405,72 +286,71 @@ export class Engine {
         });
         const outlineShaderModule = this.device.createShaderModule({
             label: "outline shaders",
-            code: /* wgsl */ `
-        struct CameraUniforms {
-          view: mat4x4f,
-          projection: mat4x4f,
-          viewPos: vec3f,
-          _padding: f32,
-        };
-
-        struct MaterialUniforms {
-          edgeColor: vec4f,
-          edgeSize: f32,
-          _padding1: f32,
-          _padding2: f32,
-          _padding3: f32,
-        };
-
-        @group(0) @binding(0) var<uniform> camera: CameraUniforms;
-        @group(0) @binding(1) var<uniform> material: MaterialUniforms;
-        @group(0) @binding(2) var<storage, read> skinMats: array<mat4x4f>;
-
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-        };
-
-        @vertex fn vs(
-          @location(0) position: vec3f,
-          @location(1) normal: vec3f,
-          @location(2) uv: vec2f,
-          @location(3) joints0: vec4<u32>,
-          @location(4) weights0: vec4<f32>
-        ) -> VertexOutput {
-          var output: VertexOutput;
-          let pos4 = vec4f(position, 1.0);
-          
-          // Normalize weights to ensure they sum to 1.0 (handles floating-point precision issues)
-          let weightSum = weights0.x + weights0.y + weights0.z + weights0.w;
-          var normalizedWeights: vec4f;
-          if (weightSum > 0.0001) {
-            normalizedWeights = weights0 / weightSum;
-          } else {
-            normalizedWeights = vec4f(1.0, 0.0, 0.0, 0.0);
-          }
-          
-          var skinnedPos = vec4f(0.0, 0.0, 0.0, 0.0);
-          var skinnedNrm = vec3f(0.0, 0.0, 0.0);
-          for (var i = 0u; i < 4u; i++) {
-            let j = joints0[i];
-            let w = normalizedWeights[i];
-            let m = skinMats[j];
-            skinnedPos += (m * pos4) * w;
-            let r3 = mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz);
-            skinnedNrm += (r3 * normal) * w;
-          }
-          let worldPos = skinnedPos.xyz;
-          let worldNormal = normalize(skinnedNrm);
-          
-          // MMD invert hull: expand vertices outward along normals
-          let scaleFactor = 0.01;
-          let expandedPos = worldPos + worldNormal * material.edgeSize * scaleFactor;
-          output.position = camera.projection * camera.view * vec4f(expandedPos, 1.0);
-          return output;
-        }
-
-        @fragment fn fs() -> @location(0) vec4f {
-          return material.edgeColor;
-        }
+            code: /* wgsl */ `
+        struct CameraUniforms {
+          view: mat4x4f,
+          projection: mat4x4f,
+          viewPos: vec3f,
+          _padding: f32,
+        };
+
+        struct MaterialUniforms {
+          edgeColor: vec4f,
+          edgeSize: f32,
+          _padding1: f32,
+          _padding2: f32,
+          _padding3: f32,
+        };
+
+        @group(0) @binding(0) var<uniform> camera: CameraUniforms;
+        @group(0) @binding(1) var<uniform> material: MaterialUniforms;
+        @group(0) @binding(2) var<storage, read> skinMats: array<mat4x4f>;
+
+        struct VertexOutput {
+          @builtin(position) position: vec4f,
+        };
+
+        @vertex fn vs(
+          @location(0) position: vec3f,
+          @location(1) normal: vec3f,
+          @location(3) joints0: vec4<u32>,
+          @location(4) weights0: vec4<f32>
+        ) -> VertexOutput {
+          var output: VertexOutput;
+          let pos4 = vec4f(position, 1.0);
+          
+          // Normalize weights to ensure they sum to 1.0 (handles floating-point precision issues)
+          let weightSum = weights0.x + weights0.y + weights0.z + weights0.w;
+          var normalizedWeights: vec4f;
+          if (weightSum > 0.0001) {
+            normalizedWeights = weights0 / weightSum;
+          } else {
+            normalizedWeights = vec4f(1.0, 0.0, 0.0, 0.0);
+          }
+          
+          var skinnedPos = vec4f(0.0, 0.0, 0.0, 0.0);
+          var skinnedNrm = vec3f(0.0, 0.0, 0.0);
+          for (var i = 0u; i < 4u; i++) {
+            let j = joints0[i];
+            let w = normalizedWeights[i];
+            let m = skinMats[j];
+            skinnedPos += (m * pos4) * w;
+            let r3 = mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz);
+            skinnedNrm += (r3 * normal) * w;
+          }
+          let worldPos = skinnedPos.xyz;
+          let worldNormal = normalize(skinnedNrm);
+          
+          // MMD invert hull: expand vertices outward along normals
+          let scaleFactor = 0.01;
+          let expandedPos = worldPos + worldNormal * material.edgeSize * scaleFactor;
+          output.position = camera.projection * camera.view * vec4f(expandedPos, 1.0);
+          return output;
+        }
+
+        @fragment fn fs() -> @location(0) vec4f {
+          return material.edgeColor;
+        }
       `,
         });
         this.outlinePipeline = this.device.createRenderPipeline({
@@ -492,11 +372,6 @@ export class Engine {
                                 offset: 3 * 4,
                                 format: "float32x3",
                             },
-                            {
-                                shaderLocation: 2,
-                                offset: 6 * 4,
-                                format: "float32x2",
-                            },
                         ],
                     },
                     {
@@ -561,11 +436,6 @@ export class Engine {
                                 offset: 3 * 4,
                                 format: "float32x3",
                             },
-                            {
-                                shaderLocation: 2,
-                                offset: 6 * 4,
-                                format: "float32x2",
-                            },
                         ],
                     },
                     {
@@ -642,11 +512,6 @@ export class Engine {
                                 offset: 3 * 4,
                                 format: "float32x3",
                             },
-                            {
-                                shaderLocation: 2,
-                                offset: 6 * 4,
-                                format: "float32x2",
-                            },
                         ],
                     },
                     {
@@ -712,7 +577,7 @@ export class Engine {
             label: "hair pipeline (over eyes)",
             layout: sharedPipelineLayout,
             vertex: {
-                module: hairShaderModule,
+                module: shaderModule,
                 buffers: [
                     {
                         arrayStride: 8 * 4,
@@ -733,7 +598,7 @@ export class Engine {
                 ],
             },
             fragment: {
-                module: hairShaderModule,
+                module: shaderModule,
                 targets: [
                     {
                         format: this.presentationFormat,
@@ -777,7 +642,7 @@ export class Engine {
             label: "hair pipeline (over non-eyes)",
             layout: sharedPipelineLayout,
             vertex: {
-                module: hairShaderModule,
+                module: shaderModule,
                 buffers: [
                     {
                         arrayStride: 8 * 4,
@@ -798,7 +663,7 @@ export class Engine {
                 ],
             },
             fragment: {
-                module: hairShaderModule,
+                module: shaderModule,
                 targets: [
                     {
                         format: this.presentationFormat,
@@ -907,31 +772,31 @@ export class Engine {
     createSkinMatrixComputePipeline() {
         const computeShader = this.device.createShaderModule({
             label: "skin matrix compute",
-            code: /* wgsl */ `
-        struct BoneCountUniform {
-          count: u32,
-          _padding1: u32,
-          _padding2: u32,
-          _padding3: u32,
-          _padding4: vec4<u32>,
-        };
-        
-        @group(0) @binding(0) var<uniform> boneCount: BoneCountUniform;
-        @group(0) @binding(1) var<storage, read> worldMatrices: array<mat4x4f>;
-        @group(0) @binding(2) var<storage, read> inverseBindMatrices: array<mat4x4f>;
-        @group(0) @binding(3) var<storage, read_write> skinMatrices: array<mat4x4f>;
-        
-        @compute @workgroup_size(64)
-        fn main(@builtin(global_invocation_id) globalId: vec3<u32>) {
-          let boneIndex = globalId.x;
-          // Bounds check: we dispatch workgroups (64 threads each), so some threads may be out of range
-          if (boneIndex >= boneCount.count) {
-            return;
-          }
-          let worldMat = worldMatrices[boneIndex];
-          let invBindMat = inverseBindMatrices[boneIndex];
-          skinMatrices[boneIndex] = worldMat * invBindMat;
-        }
+            code: /* wgsl */ `
+        struct BoneCountUniform {
+          count: u32,
+          _padding1: u32,
+          _padding2: u32,
+          _padding3: u32,
+          _padding4: vec4<u32>,
+        };
+        
+        @group(0) @binding(0) var<uniform> boneCount: BoneCountUniform;
+        @group(0) @binding(1) var<storage, read> worldMatrices: array<mat4x4f>;
+        @group(0) @binding(2) var<storage, read> inverseBindMatrices: array<mat4x4f>;
+        @group(0) @binding(3) var<storage, read_write> skinMatrices: array<mat4x4f>;
+        
+        @compute @workgroup_size(64)
+        fn main(@builtin(global_invocation_id) globalId: vec3<u32>) {
+          let boneIndex = globalId.x;
+          // Bounds check: we dispatch workgroups (64 threads each), so some threads may be out of range
+          if (boneIndex >= boneCount.count) {
+            return;
+          }
+          let worldMat = worldMatrices[boneIndex];
+          let invBindMat = inverseBindMatrices[boneIndex];
+          skinMatrices[boneIndex] = worldMat * invBindMat;
+        }
       `,
         });
         this.skinMatrixComputePipeline = this.device.createComputePipeline({
@@ -985,143 +850,143 @@ export class Engine {
         // Bloom extraction shader (extracts bright areas)
         const bloomExtractShader = this.device.createShaderModule({
             label: "bloom extract",
-            code: /* wgsl */ `
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-          @location(0) uv: vec2f,
-        };
-
-        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
-          var output: VertexOutput;
-          // Generate fullscreen quad from vertex index
-          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
-          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
-          output.position = vec4f(x, y, 0.0, 1.0);
-          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
-          return output;
-        }
-
-        struct BloomExtractUniforms {
-          threshold: f32,
-          _padding1: f32,
-          _padding2: f32,
-          _padding3: f32,
-          _padding4: f32,
-          _padding5: f32,
-          _padding6: f32,
-          _padding7: f32,
-        };
-
-        @group(0) @binding(0) var inputTexture: texture_2d<f32>;
-        @group(0) @binding(1) var inputSampler: sampler;
-        @group(0) @binding(2) var<uniform> extractUniforms: BloomExtractUniforms;
-
-        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
-          let color = textureSample(inputTexture, inputSampler, input.uv);
-          // Extract bright areas above threshold
-          let threshold = extractUniforms.threshold;
-          let bloom = max(vec3f(0.0), color.rgb - vec3f(threshold)) / max(0.001, 1.0 - threshold);
-          return vec4f(bloom, color.a);
-        }
+            code: /* wgsl */ `
+        struct VertexOutput {
+          @builtin(position) position: vec4f,
+          @location(0) uv: vec2f,
+        };
+
+        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
+          var output: VertexOutput;
+          // Generate fullscreen quad from vertex index
+          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
+          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
+          output.position = vec4f(x, y, 0.0, 1.0);
+          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
+          return output;
+        }
+
+        struct BloomExtractUniforms {
+          threshold: f32,
+          _padding1: f32,
+          _padding2: f32,
+          _padding3: f32,
+          _padding4: f32,
+          _padding5: f32,
+          _padding6: f32,
+          _padding7: f32,
+        };
+
+        @group(0) @binding(0) var inputTexture: texture_2d<f32>;
+        @group(0) @binding(1) var inputSampler: sampler;
+        @group(0) @binding(2) var<uniform> extractUniforms: BloomExtractUniforms;
+
+        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+          let color = textureSample(inputTexture, inputSampler, input.uv);
+          // Extract bright areas above threshold
+          let threshold = extractUniforms.threshold;
+          let bloom = max(vec3f(0.0), color.rgb - vec3f(threshold)) / max(0.001, 1.0 - threshold);
+          return vec4f(bloom, color.a);
+        }
       `,
         });
         // Bloom blur shader (gaussian blur - can be used for both horizontal and vertical)
         const bloomBlurShader = this.device.createShaderModule({
             label: "bloom blur",
-            code: /* wgsl */ `
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-          @location(0) uv: vec2f,
-        };
-
-        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
-          var output: VertexOutput;
-          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
-          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
-          output.position = vec4f(x, y, 0.0, 1.0);
-          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
-          return output;
-        }
-
-        struct BlurUniforms {
-          direction: vec2f,
-          _padding1: f32,
-          _padding2: f32,
-          _padding3: f32,
-          _padding4: f32,
-          _padding5: f32,
-          _padding6: f32,
-        };
-
-        @group(0) @binding(0) var inputTexture: texture_2d<f32>;
-        @group(0) @binding(1) var inputSampler: sampler;
-        @group(0) @binding(2) var<uniform> blurUniforms: BlurUniforms;
-
-        // 9-tap gaussian blur
-        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
-          let texelSize = 1.0 / vec2f(textureDimensions(inputTexture));
-          var result = vec4f(0.0);
-          
-          // Gaussian weights for 9-tap filter
-          let weights = array<f32, 9>(
-            0.01621622, 0.05405405, 0.12162162,
-            0.19459459, 0.22702703,
-            0.19459459, 0.12162162, 0.05405405, 0.01621622
-          );
-          
-          let offsets = array<f32, 9>(-4.0, -3.0, -2.0, -1.0, 0.0, 1.0, 2.0, 3.0, 4.0);
-          
-          for (var i = 0u; i < 9u; i++) {
-            let offset = offsets[i] * texelSize * blurUniforms.direction;
-            result += textureSample(inputTexture, inputSampler, input.uv + offset) * weights[i];
-          }
-          
-          return result;
-        }
+            code: /* wgsl */ `
+        struct VertexOutput {
+          @builtin(position) position: vec4f,
+          @location(0) uv: vec2f,
+        };
+
+        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
+          var output: VertexOutput;
+          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
+          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
+          output.position = vec4f(x, y, 0.0, 1.0);
+          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
+          return output;
+        }
+
+        struct BlurUniforms {
+          direction: vec2f,
+          _padding1: f32,
+          _padding2: f32,
+          _padding3: f32,
+          _padding4: f32,
+          _padding5: f32,
+          _padding6: f32,
+        };
+
+        @group(0) @binding(0) var inputTexture: texture_2d<f32>;
+        @group(0) @binding(1) var inputSampler: sampler;
+        @group(0) @binding(2) var<uniform> blurUniforms: BlurUniforms;
+
+        // 9-tap gaussian blur
+        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+          let texelSize = 1.0 / vec2f(textureDimensions(inputTexture));
+          var result = vec4f(0.0);
+          
+          // Gaussian weights for 9-tap filter
+          let weights = array<f32, 9>(
+            0.01621622, 0.05405405, 0.12162162,
+            0.19459459, 0.22702703,
+            0.19459459, 0.12162162, 0.05405405, 0.01621622
+          );
+          
+          let offsets = array<f32, 9>(-4.0, -3.0, -2.0, -1.0, 0.0, 1.0, 2.0, 3.0, 4.0);
+          
+          for (var i = 0u; i < 9u; i++) {
+            let offset = offsets[i] * texelSize * blurUniforms.direction;
+            result += textureSample(inputTexture, inputSampler, input.uv + offset) * weights[i];
+          }
+          
+          return result;
+        }
       `,
         });
         // Bloom composition shader (combines original scene with bloom)
         const bloomComposeShader = this.device.createShaderModule({
             label: "bloom compose",
-            code: /* wgsl */ `
-        struct VertexOutput {
-          @builtin(position) position: vec4f,
-          @location(0) uv: vec2f,
-        };
-
-        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
-          var output: VertexOutput;
-          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
-          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
-          output.position = vec4f(x, y, 0.0, 1.0);
-          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
-          return output;
-        }
-
-        struct BloomComposeUniforms {
-          intensity: f32,
-          _padding1: f32,
-          _padding2: f32,
-          _padding3: f32,
-          _padding4: f32,
-          _padding5: f32,
-          _padding6: f32,
-          _padding7: f32,
-        };
-
-        @group(0) @binding(0) var sceneTexture: texture_2d<f32>;
-        @group(0) @binding(1) var sceneSampler: sampler;
-        @group(0) @binding(2) var bloomTexture: texture_2d<f32>;
-        @group(0) @binding(3) var bloomSampler: sampler;
-        @group(0) @binding(4) var<uniform> composeUniforms: BloomComposeUniforms;
-
-        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
-          let scene = textureSample(sceneTexture, sceneSampler, input.uv);
-          let bloom = textureSample(bloomTexture, bloomSampler, input.uv);
-          // Additive blending with intensity control
-          let result = scene.rgb + bloom.rgb * composeUniforms.intensity;
-          return vec4f(result, scene.a);
-        }
+            code: /* wgsl */ `
+        struct VertexOutput {
+          @builtin(position) position: vec4f,
+          @location(0) uv: vec2f,
+        };
+
+        @vertex fn vs(@builtin(vertex_index) vertexIndex: u32) -> VertexOutput {
+          var output: VertexOutput;
+          let x = f32((vertexIndex << 1u) & 2u) * 2.0 - 1.0;
+          let y = f32(vertexIndex & 2u) * 2.0 - 1.0;
+          output.position = vec4f(x, y, 0.0, 1.0);
+          output.uv = vec2f(x * 0.5 + 0.5, 1.0 - (y * 0.5 + 0.5));
+          return output;
+        }
+
+        struct BloomComposeUniforms {
+          intensity: f32,
+          _padding1: f32,
+          _padding2: f32,
+          _padding3: f32,
+          _padding4: f32,
+          _padding5: f32,
+          _padding6: f32,
+          _padding7: f32,
+        };
+
+        @group(0) @binding(0) var sceneTexture: texture_2d<f32>;
+        @group(0) @binding(1) var sceneSampler: sampler;
+        @group(0) @binding(2) var bloomTexture: texture_2d<f32>;
+        @group(0) @binding(3) var bloomSampler: sampler;
+        @group(0) @binding(4) var<uniform> composeUniforms: BloomComposeUniforms;
+
+        @fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+          let scene = textureSample(sceneTexture, sceneSampler, input.uv);
+          let bloom = textureSample(bloomTexture, bloomSampler, input.uv);
+          // Additive blending with intensity control
+          let result = scene.rgb + bloom.rgb * composeUniforms.intensity;
+          return vec4f(result, scene.a);
+        }
       `,
         });
         // Create uniform buffer for blur direction (minimum 32 bytes for WebGPU)
@@ -1520,17 +1385,16 @@ export class Engine {
             const materialAlpha = mat.diffuse[3];
             const EPSILON = 0.001;
             const isTransparent = materialAlpha < 1.0 - EPSILON;
-            // Create material uniform data - for hair materials, we'll create two versions
-            // MaterialUniforms struct: alpha, rimIntensity, rimPower, _padding1, rimColor (vec3), _padding2
+            // Create material uniform data
             const materialUniformData = new Float32Array(8);
             materialUniformData[0] = materialAlpha;
-            materialUniformData[1] = this.rimLightIntensity;
-            materialUniformData[2] = this.rimLightPower;
-            materialUniformData[3] = 0.0; // _padding1
+            materialUniformData[1] = 1.0; // alphaMultiplier: 1.0 for non-hair materials
+            materialUniformData[2] = this.rimLightIntensity;
+            materialUniformData[3] = this.rimLightPower;
             materialUniformData[4] = this.rimLightColor[0]; // rimColor.r
             materialUniformData[5] = this.rimLightColor[1]; // rimColor.g
             materialUniformData[6] = this.rimLightColor[2]; // rimColor.b
-            materialUniformData[7] = 0.0; // _padding2
+            materialUniformData[7] = 0.0; // _padding1
             const materialUniformBuffer = this.device.createBuffer({
                 label: `material uniform: ${mat.name}`,
                 size: materialUniformData.byteLength,
@@ -1563,7 +1427,6 @@ export class Engine {
             }
             else if (mat.isHair) {
                 // For hair materials, create two bind groups: one for over-eyes (alphaMultiplier = 0.5) and one for over-non-eyes (alphaMultiplier = 1.0)
-                // Hair MaterialUniforms struct: alpha, alphaMultiplier, rimIntensity, rimPower, rimColor (vec3), _padding1
                 const materialUniformDataOverEyes = new Float32Array(8);
                 materialUniformDataOverEyes[0] = materialAlpha;
                 materialUniformDataOverEyes[1] = 0.5; // alphaMultiplier: 0.5 for over-eyes
@@ -2010,16 +1873,17 @@ export class Engine {
     }
     // Update model pose and physics
     updateModelPose(deltaTime) {
+        // Step 1: Animation evaluation (computes matrices to CPU memory, no upload yet)
         this.currentModel.evaluatePose();
-        // Upload world matrices to GPU
+        // Step 2: Get world matrices (still in CPU memory)
         const worldMats = this.currentModel.getBoneWorldMatrices();
-        this.device.queue.writeBuffer(this.worldMatrixBuffer, 0, worldMats.buffer, worldMats.byteOffset, worldMats.byteLength);
+        // Step 3: Physics modifies matrices in-place
         if (this.physics) {
             this.physics.step(deltaTime, worldMats, this.currentModel.getBoneInverseBindMatrices());
-            // Re-upload world matrices after physics (physics may have updated bones)
-            this.device.queue.writeBuffer(this.worldMatrixBuffer, 0, worldMats.buffer, worldMats.byteOffset, worldMats.byteLength);
         }
-        // Compute skin matrices on GPU
+        // Step 4: Upload ONCE with final result (animation + physics)
+        this.device.queue.writeBuffer(this.worldMatrixBuffer, 0, worldMats.buffer, worldMats.byteOffset, worldMats.byteLength);
+        // Step 5: GPU skinning
         this.computeSkinMatrices();
     }
     // Compute skin matrices on GPU
@@ -2123,20 +1987,56 @@ export class Engine {
             if (skeleton)
                 bufferMemoryBytes += Math.max(256, skeleton.bones.length * 16 * 4);
         }
+        if (this.worldMatrixBuffer) {
+            const skeleton = this.currentModel?.getSkeleton();
+            if (skeleton)
+                bufferMemoryBytes += Math.max(256, skeleton.bones.length * 16 * 4);
+        }
+        if (this.inverseBindMatrixBuffer) {
+            const skeleton = this.currentModel?.getSkeleton();
+            if (skeleton)
+                bufferMemoryBytes += Math.max(256, skeleton.bones.length * 16 * 4);
+        }
         bufferMemoryBytes += 40 * 4; // cameraUniformBuffer
         bufferMemoryBytes += 64 * 4; // lightUniformBuffer
+        bufferMemoryBytes += 32; // boneCountBuffer
+        bufferMemoryBytes += 32; // blurDirectionBuffer
+        bufferMemoryBytes += 32; // bloomIntensityBuffer
+        bufferMemoryBytes += 32; // bloomThresholdBuffer
+        if (this.fullscreenQuadBuffer) {
+            bufferMemoryBytes += 24 * 4; // fullscreenQuadBuffer (6 vertices * 4 floats)
+        }
+        // Material uniform buffers: Float32Array(8) = 32 bytes each
         const totalMaterialDraws = this.opaqueNonEyeNonHairDraws.length +
             this.eyeDraws.length +
             this.hairDrawsOverEyes.length +
             this.hairDrawsOverNonEyes.length +
             this.transparentNonEyeNonHairDraws.length;
-        bufferMemoryBytes += totalMaterialDraws * 4; // Material uniform buffers
+        bufferMemoryBytes += totalMaterialDraws * 32; // Material uniform buffers (8 floats = 32 bytes)
+        // Outline material uniform buffers: Float32Array(8) = 32 bytes each
+        const totalOutlineDraws = this.opaqueNonEyeNonHairOutlineDraws.length +
+            this.eyeOutlineDraws.length +
+            this.hairOutlineDraws.length +
+            this.transparentNonEyeNonHairOutlineDraws.length;
+        bufferMemoryBytes += totalOutlineDraws * 32; // Outline material uniform buffers
         let renderTargetMemoryBytes = 0;
         if (this.multisampleTexture) {
             const width = this.canvas.width;
             const height = this.canvas.height;
             renderTargetMemoryBytes += width * height * 4 * this.sampleCount; // multisample color
-            renderTargetMemoryBytes += width * height * 4; // depth
+            renderTargetMemoryBytes += width * height * 4; // depth (depth24plus-stencil8 = 4 bytes)
+        }
+        if (this.sceneRenderTexture) {
+            const width = this.canvas.width;
+            const height = this.canvas.height;
+            renderTargetMemoryBytes += width * height * 4; // sceneRenderTexture (non-multisampled)
+        }
+        if (this.bloomExtractTexture) {
+            const width = Math.floor(this.canvas.width / 2);
+            const height = Math.floor(this.canvas.height / 2);
+            renderTargetMemoryBytes += width * height * 4; // bloomExtractTexture
+            renderTargetMemoryBytes += width * height * 4; // bloomBlurTexture1
+            renderTargetMemoryBytes += width * height * 4; // bloomBlurTexture2
         }
         const totalGPUMemoryBytes = textureMemoryBytes + bufferMemoryBytes + renderTargetMemoryBytes;
         this.stats.gpuMemory = Math.round((totalGPUMemoryBytes / 1024 / 1024) * 100) / 100;