Faster default form in the root; the first release's form moves to 262k/

Two forms. The repo's root now holds a faster default form: a photo makes a splat of 65,536 Gaussians (2,048 anchors), every graph but BiRefNet runs on the GPU at FP16, the denoiser's joint blocks 1-22 share one graph (den_joint_mid.tflite, weights in den_joint_w.f32) and the Gaussian decoder's graphs 1-6 share another (gs_mid.tflite, weights in gs_w.f32); the app copies a block's weights in before each call, so each is compiled once per photo. The first release's form (262,144 Gaussians, a graph per block, GPU FP16 with fp32 accumulation) moves to 262k/ with its own manifest.json; its files are the same bytes as the root's at commit e7ee1226. On a Galaxy S26 the first photo after start on a cool phone took 158 s with the default form and 267 s with the 262k/ form. Four-view renders against the author's default on the phone (armchair / axolotl / succulent; threshold 25 dB): 34.41 / 32.26 / 34.26 dB PSNR with the default form, 46.59 / 42.34 / 41.43 dB with the 262k/ form.

.gitattributes

@@ -35,3 +35,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 den_pos_emb.f32 filter=lfs diff=lfs merge=lfs -text
 assets/hero.png filter=lfs diff=lfs merge=lfs -text
+den_joint_w.f32 filter=lfs diff=lfs merge=lfs -text
+gs_w.f32 filter=lfs diff=lfs merge=lfs -text

262k/bir.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42b98bf0e497dd596327c31883fec21fe7013a4c9f72d1f67ec8ff4d3f0dfd43
+size 930410016

262k/den_ctx.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39278b27d00a3dcc70da5e2737f6055d5ba4e7e817423a4cf176aacbfb86230d
+size 117711308

262k/den_in.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1350c12da702a4a0abf53cdf528b62a8a75ea4d948c6a4764b264b927b419802
+size 146782524

262k/den_joint_00.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37e269459ef1325bd4a84d1df1545afacf8e33d6d5c232c62f834f343a882f86
+size 56134468

262k/den_joint_23.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53420d4ff17eeee40d89480fecc549d578b8877a386821eb76c2754a4737ebdb
+size 56235684

262k/den_pos_emb.f32

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1dd2858023be2f47f573ee856d4f0ed2328d47fe0f88d3984038a1b1ffeabb2
+size 33554432

262k/gs_0.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:244ae2ffc6d510c8eab99d2381cb20edb8a7c1715272f7a7efeba5fcd77cfa39
+size 122497768

262k/gs_7.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef75898f78654d4d10692b333dc1e12c860ee54138795aece8e2306ba76217b7
+size 120252780

262k/gs_rep.json

@@ -0,0 +1,17 @@
+{
+ "base_offset_scale": -2.970628499984741,
+ "rep": {
+  "num_gaussians": 32,
+  "lr": {
+   "_xyz": 1.0,
+   "_features_dc": 1.0,
+   "_opacity": 1.0,
+   "_scaling": 1.0,
+   "_rotation": 0.1
+  },
+  "perturbe_size": 1.5,
+  "3d_filter_kernel_size": 0.0009,
+  "scaling_bias": 0.004,
+  "opacity_bias": 0.1
+ }
+}

262k/manifest.json

@@ -0,0 +1,2316 @@
+{
+ "matting": {
+  "default": "bir",
+  "alternative": "dis"
+ },
+ "graphs": {
+  "bir": {
+   "file": "bir.tflite",
+   "bytes": 930410016,
+   "accelerator": "cpu",
+   "precision": "fp32",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        3,
+        1024,
+        1024
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        1,
+        1024,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "42b98bf0e497dd596327c31883fec21fe7013a4c9f72d1f67ec8ff4d3f0dfd43",
+   "run_peak_bytes": 2798186496
+  },
+  "dis": {
+   "file": "dis.tflite",
+   "bytes": 176223448,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        3,
+        1024,
+        1024
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        1,
+        1024,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "0c3c93b6a2a65e7c69137ec82596944e6bfb97d982c75bab03acb0738dbaa087",
+   "repo": "litert-community/DIS-ISNet-LiteRT"
+  },
+  "dino_0": {
+   "file": "dino_0.tflite",
+   "bytes": 843659604,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        3,
+        1024,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "7b4fc27cde00f794a846766b660445b08cb94e2380ad9c7f58db191edd3b5b50",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "dino_1": {
+   "file": "dino_1.tflite",
+   "bytes": 839694588,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "54a96f0a0ac0cb29e3520c303dba10544ab88f67d97160dbae941ea8f09b41be",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "dino_2": {
+   "file": "dino_2.tflite",
+   "bytes": 839694588,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "93b8659dfe738efb3239aa28ff683d503f4caee7b73ebbb9e8914ae27e90f3a1",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "dino_3": {
+   "file": "dino_3.tflite",
+   "bytes": 839710040,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1,
+        4101,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "9281d490c988268cd8209f353937677b585e13e6ea324c4f01229f93c89f8c39",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "vae_0": {
+   "file": "vae_0.tflite",
+   "bytes": 3036028,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        3,
+        1024,
+        1024
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        128,
+        512,
+        512
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "a8c8b881165717d854bead656620f77bfccc71b5f7772c160753723c5f80c2d7"
+  },
+  "vae_1": {
+   "file": "vae_1.tflite",
+   "bytes": 10858720,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        128,
+        512,
+        512
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        256,
+        256,
+        256
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "07e357358a2e50a14462be77433f0c11ae2319d0912d320a65335d71fabfd48b"
+  },
+  "vae_2": {
+   "file": "vae_2.tflite",
+   "bytes": 43164896,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        256,
+        256,
+        256
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "203175990bbbf43a66263ae69ec8000f19bc943314226bf2e26b3d0b0b26eb3f"
+  },
+  "vae_3": {
+   "file": "vae_3.tflite",
+   "bytes": 56792512,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "d3375079a1558cdd4f4c3f96337830ce81d2f37089c2cd79e701cfc393ecc67e"
+  },
+  "vae_attn": {
+   "file": "vae_attn.tflite",
+   "bytes": 4423776,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "4bec7ec37ad58661abb63ee1e16ccb7646afc6d617acf3caa714721912237d88"
+  },
+  "vae_4": {
+   "file": "vae_4.tflite",
+   "bytes": 20179348,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        512,
+        128,
+        128
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        64,
+        128,
+        128
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "6ea00886e1fb2bebbd4be40b6591a04abea2208fd6d59e32a4d52310b0d35853"
+  },
+  "den_ctx": {
+   "file": "den_ctx.tflite",
+   "bytes": 117711308,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        4101,
+        1280
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        4101,
+        128
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4101,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "39278b27d00a3dcc70da5e2737f6055d5ba4e7e817423a4cf176aacbfb86230d"
+  },
+  "den_in": {
+   "file": "den_in.tflite",
+   "bytes": 146782524,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        16
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1,
+        5
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_4",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "output_1",
+       [
+        1,
+        1,
+        1024
+       ]
+      ],
+      [
+       "output_2",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "output_3",
+       [
+        1,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "1350c12da702a4a0abf53cdf528b62a8a75ea4d948c6a4764b264b927b419802"
+  },
+  "den_joint_00": {
+   "file": "den_joint_00.tflite",
+   "bytes": 56134468,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        4101,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1,
+        1024
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_4",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "37e269459ef1325bd4a84d1df1545afacf8e33d6d5c232c62f834f343a882f86"
+  },
+  "den_joint_01": {
+   "file": "den_joint_01.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "103efe94ce85aeb29f084f85962d0e85d4a395c629101d3ea21d23e6cdc53e25"
+  },
+  "den_joint_02": {
+   "file": "den_joint_02.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "1ec688cf610156841a1523db9ed93ad268274f957132dcbc92539308143ce517"
+  },
+  "den_joint_03": {
+   "file": "den_joint_03.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "b5828ac85b0f42118e79f2e83cbe05021d2b68f9a3c167f0ac014a3026799cc2"
+  },
+  "den_joint_04": {
+   "file": "den_joint_04.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "ea5c57b89c8d8bbc5144d5d487e20fc49daf25308bccf853bb3c9d4581ae33fe"
+  },
+  "den_joint_05": {
+   "file": "den_joint_05.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "f339d8b79e5d4c2fc408c2d42f7ce959b81821cad37861b2a1294b8397e19997"
+  },
+  "den_joint_06": {
+   "file": "den_joint_06.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "30d97e62ac4fc28882903961c88e65e6fe540d61e95504651b44d7cb8f341c4c"
+  },
+  "den_joint_07": {
+   "file": "den_joint_07.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "0c7b243e55c69b6cdd9c5e7bbef313f6cc6be0c0f7ad32e428187e22b2183a67"
+  },
+  "den_joint_08": {
+   "file": "den_joint_08.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "0eb17bea8ae04ea0d88d79b327515e1a106bb5339d72440a8ec63712865e3f27"
+  },
+  "den_joint_09": {
+   "file": "den_joint_09.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "353ece07e6b4888d80ab08e2bdd6dc2dd456f213ca87694f6d77f987d5d4debc"
+  },
+  "den_joint_10": {
+   "file": "den_joint_10.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "0c7e15340ea4212369ae85f3f9db625f21ccd1747038e0725675c5a7ca32ac07"
+  },
+  "den_joint_11": {
+   "file": "den_joint_11.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "76cc264ae1ce820749f9dd4758e07fa548ee805a7487cbd4ba342eccc523d779"
+  },
+  "den_joint_12": {
+   "file": "den_joint_12.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "75ff657c9c7cca82e45ec36fc259cde4cfc90829f09639cdc749d8bddce73f27"
+  },
+  "den_joint_13": {
+   "file": "den_joint_13.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "9ecae224a9b4962e1e1b1d719942ea89182c77b1f1cf629017f7117d9dded5b2"
+  },
+  "den_joint_14": {
+   "file": "den_joint_14.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "59f957bdcbcef54f90ac90fc2ed9ff52f40fa93afeb589f6fa377c734ddb4c5c"
+  },
+  "den_joint_15": {
+   "file": "den_joint_15.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "4839713f18df82b2dd53ed2d605a9d98240bee6bebc00e281a1316246b5ffaaa"
+  },
+  "den_joint_16": {
+   "file": "den_joint_16.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "a8cc4d07fe9830a8f2fc72420db955c6d08f6093fc82ea9b329df4d3275599e2"
+  },
+  "den_joint_17": {
+   "file": "den_joint_17.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "29c8966f6b967250b818ebc902b0b1c6070436700c4cd52ad65ab26a3b102387"
+  },
+  "den_joint_18": {
+   "file": "den_joint_18.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "e5f035dbfa13fcbb92f0212d4c773b8595ae1765b6658a9ecac11dfc80059b21"
+  },
+  "den_joint_19": {
+   "file": "den_joint_19.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "48394c1fbb680f7bb68ccf3e0253afc170ef47f0f14dc1163d0c84fb9c46222d"
+  },
+  "den_joint_20": {
+   "file": "den_joint_20.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "2be313d01a72354c1b5dbc291ff1dcecf6345b02e7805e9214c019094762d063"
+  },
+  "den_joint_21": {
+   "file": "den_joint_21.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "c9c05edb15046307eb06de930d335bdc3b532f5eb9a79314664bb6ecfb087e4e"
+  },
+  "den_joint_22": {
+   "file": "den_joint_22.tflite",
+   "bytes": 56134100,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "e7674dcda5bef2fca961742e0cb58736e9bed70485a2cad96080bbcddea3490b"
+  },
+  "den_joint_23": {
+   "file": "den_joint_23.tflite",
+   "bytes": 56235684,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        12294,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        6144
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        1024
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        64
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        16
+       ]
+      ],
+      [
+       "output_1",
+       [
+        1,
+        1,
+        5
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "53420d4ff17eeee40d89480fecc549d578b8877a386821eb76c2754a4737ebdb"
+  },
+  "oct": {
+   "file": "oct.tflite",
+   "bytes": 204247600,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "L64": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        64,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        64,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        64,
+        8
+       ]
+      ]
+     ]
+    },
+    "L256": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        256,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        256,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        256,
+        8
+       ]
+      ]
+     ]
+    },
+    "L1024": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        1024,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        1024,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        1024,
+        8
+       ]
+      ]
+     ]
+    },
+    "L4096": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        4096,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        4096,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        4096,
+        8
+       ]
+      ]
+     ]
+    },
+    "L8192": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        256
+       ]
+      ],
+      [
+       "args_3",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        8
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "6dab2c2dc79013e798ffabf41dea246bf447cee7bc9336430e8d773162a7cda7"
+  },
+  "gs_0": {
+   "file": "gs_0.tflite",
+   "bytes": 122497768,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        3
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_2",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "244ae2ffc6d510c8eab99d2381cb20edb8a7c1715272f7a7efeba5fcd77cfa39"
+  },
+  "gs_1": {
+   "file": "gs_1.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "08e405e5b545d61b02b9f053d79c09f1300b57cd0f4469ffa65853eb8178bd4d"
+  },
+  "gs_2": {
+   "file": "gs_2.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "4280e07069d2a6cd271e54477164b8b632d588030e46909a1854a8b35e30eb93"
+  },
+  "gs_3": {
+   "file": "gs_3.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "5dcd7ecb298a160af606530df20f018b91384cd25e06fe25a0b42401cba7e651"
+  },
+  "gs_4": {
+   "file": "gs_4.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "37541cee18d0de7b62e104391af8f61eac192424f58a9613cd8281fe6c23f472"
+  },
+  "gs_5": {
+   "file": "gs_5.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "80972de6ec50f8d6f1ace44c47f076d0f214c0cdb99b6467d37e82af1de940c3"
+  },
+  "gs_6": {
+   "file": "gs_6.tflite",
+   "bytes": 118281876,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "50dcc38afa5253d24fb792798db27327905b766404506405ead8e525cb0ec177"
+  },
+  "gs_7": {
+   "file": "gs_7.tflite",
+   "bytes": 120252780,
+   "accelerator": "gpu",
+   "precision": "fp16_with_fp32_accum",
+   "signatures": {
+    "serving_default": {
+     "inputs": [
+      [
+       "args_0",
+       [
+        1,
+        8192,
+        1024
+       ]
+      ],
+      [
+       "args_1",
+       [
+        1,
+        8192,
+        16
+       ]
+      ]
+     ],
+     "outputs": [
+      [
+       "output_0",
+       [
+        1,
+        8192,
+        480
+       ]
+      ]
+     ]
+    }
+   },
+   "sha256": "ef75898f78654d4d10692b333dc1e12c860ee54138795aece8e2306ba76217b7"
+  }
+ },
+ "side_files": {
+  "den_pos_emb.f32": {
+   "shape": [
+    1,
+    8192,
+    1024
+   ],
+   "dtype": "float32_le"
+  },
+  "dino_rope_cos.f32": {
+   "shape": [
+    1,
+    1,
+    4101,
+    64
+   ],
+   "dtype": "float32_le",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "dino_rope_sin.f32": {
+   "shape": [
+    1,
+    1,
+    4101,
+    64
+   ],
+   "dtype": "float32_le",
+   "repo": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT"
+  },
+  "gs_perturbation.f32": {
+   "shape": [
+    32,
+    3
+   ],
+   "dtype": "float32_le"
+  }
+ },
+ "json_files": [
+  "vae_bn.json",
+  "gs_rep.json"
+ ],
+ "total_graph_bytes": 3837368884,
+ "repos": {
+  "main": "litert-community/Idol",
+  "dino": "litert-community/DINOv3-ViT-H16plus-Idol-LiteRT",
+  "dis": "litert-community/DIS-ISNet-LiteRT"
+ }
+}

262k/oct.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dab2c2dc79013e798ffabf41dea246bf447cee7bc9336430e8d773162a7cda7
+size 204247600

262k/vae_0.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8c8b881165717d854bead656620f77bfccc71b5f7772c160753723c5f80c2d7
+size 3036028

262k/vae_1.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07e357358a2e50a14462be77433f0c11ae2319d0912d320a65335d71fabfd48b
+size 10858720

262k/vae_2.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:203175990bbbf43a66263ae69ec8000f19bc943314226bf2e26b3d0b0b26eb3f
+size 43164896

262k/vae_3.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3375079a1558cdd4f4c3f96337830ce81d2f37089c2cd79e701cfc393ecc67e
+size 56792512

262k/vae_4.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ea00886e1fb2bebbd4be40b6591a04abea2208fd6d59e32a4d52310b0d35853
+size 20179348

262k/vae_attn.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bec7ec37ad58661abb63ee1e16ccb7646afc6d617acf3caa714721912237d88
+size 4423776

262k/vae_bn.json

@@ -0,0 +1 @@
+{"bn_mean": [-0.0673828125, -0.0712890625, -0.0751953125, -0.07470703125, 0.0223388671875, 0.0179443359375, 0.01422119140625, 0.018310546875, -6.29425048828125e-05, -0.006256103515625, -0.000209808349609375, -0.003143310546875, -0.0272216796875, -0.028076171875, -0.027587890625, -0.029052734375, -0.07666015625, -0.0673828125, -0.09033203125, -0.08935546875, 0.016845703125, 0.01519775390625, 0.00787353515625, 0.00860595703125, 0.00836181640625, 0.0015411376953125, 0.0002574920654296875, -0.0042724609375, -0.0439453125, -0.0419921875, -0.043701171875, -0.043212890625, -0.01025390625, -0.01318359375, -0.006622314453125, -0.0047607421875, -0.031005859375, -0.030517578125, -0.0279541015625, -0.0179443359375, 0.003021240234375, 0.00150299072265625, 0.0125732421875, 0.01446533203125, 0.03466796875, 0.03369140625, 0.03369140625, 0.0283203125, 0.001983642578125, 0.004730224609375, 0.004669189453125, 0.004974365234375, 0.01226806640625, 0.00811767578125, 0.008056640625, 0.01458740234375, 0.06787109375, 0.06787109375, 0.07666015625, 0.0732421875, -0.046142578125, -0.04736328125, -0.039306640625, -0.051025390625, -0.052734375, -0.0478515625, -0.047119140625, -0.0517578125, -0.03173828125, -0.03173828125, -0.034423828125, -0.0281982421875, 0.051025390625, 0.04443359375, 0.057861328125, 0.0458984375, -0.041259765625, -0.0458984375, -0.048828125, -0.046630859375, -0.00885009765625, -0.0106201171875, -0.0087890625, -0.004608154296875, -0.03759765625, -0.043212890625, -0.04345703125, -0.0498046875, 0.0118408203125, 0.0166015625, 0.020263671875, 0.0279541015625, 0.01129150390625, 0.01287841796875, 0.001556396484375, 0.00714111328125, -0.01177978515625, -0.00183868408203125, -0.01416015625, -0.00537109375, -0.00909423828125, -0.0137939453125, -0.01446533203125, -0.0186767578125, 0.0322265625, 0.030517578125, 0.02587890625, 0.0299072265625, 0.053955078125, 0.0615234375, 0.049560546875, 0.05908203125, -0.051025390625, -0.060302734375, -0.0478515625, -0.052490234375, -0.022705078125, -0.0274658203125, -0.015380859375, -0.0255126953125, -0.05712890625, -0.056396484375, -0.0517578125, -0.049560546875, 0.0115966796875, 0.00543212890625, 0.016357421875, 0.0103759765625], "bn_var": [3.25, 3.15625, 3.1875, 3.1875, 3.140625, 3.09375, 3.09375, 3.0625, 3.0, 3.015625, 3.0, 3.0, 3.46875, 3.4375, 3.46875, 3.453125, 3.09375, 3.078125, 3.078125, 3.09375, 3.015625, 3.03125, 3.015625, 3.03125, 3.0, 2.96875, 3.03125, 3.0, 3.078125, 3.0625, 3.078125, 3.0625, 3.40625, 3.40625, 3.4375, 3.4375, 3.328125, 3.171875, 3.1875, 3.21875, 3.25, 3.203125, 3.125, 3.125, 3.15625, 3.125, 3.21875, 3.1875, 3.09375, 3.03125, 3.0625, 3.015625, 3.21875, 3.234375, 3.21875, 3.21875, 3.15625, 3.15625, 3.140625, 3.15625, 3.078125, 3.046875, 3.125, 3.0625, 3.15625, 3.140625, 3.171875, 3.171875, 3.296875, 3.25, 3.25, 3.25, 3.078125, 3.0, 3.078125, 3.0625, 3.09375, 3.0625, 3.125, 3.109375, 3.125, 3.078125, 3.171875, 3.140625, 3.203125, 3.234375, 3.1875, 3.15625, 3.109375, 3.109375, 3.078125, 3.09375, 3.15625, 3.125, 3.171875, 3.171875, 2.953125, 2.90625, 2.984375, 2.921875, 3.171875, 3.09375, 3.0625, 3.046875, 3.09375, 3.0625, 3.078125, 3.015625, 3.109375, 3.09375, 3.046875, 3.03125, 3.0625, 3.09375, 3.109375, 3.109375, 2.984375, 2.9375, 3.0, 2.96875, 3.125, 3.109375, 3.140625, 3.125, 3.046875, 3.046875, 3.09375, 3.078125], "bn_eps": 0.0001}

README.md

@@ -16,10 +16,10 @@ tags:
 
 # Idol on LiteRT: one photo to a 3D Gaussian splat on the phone
 
-Idol (speridlabs/idol with the idol-turbo LoRA) as LiteRT graphs for `CompiledModel`, converted with litert-torch: on a Galaxy S26, the Android sample in `android/` turns one photo into a 3D Gaussian splat of 262,144 Gaussians in 303 s (the first photo after start on a cool phone, GPU fp16 with fp32 accumulation, LiteRT 2.3.0, measured 2026-11-09).
+Idol (speridlabs/idol with the idol-turbo LoRA) as LiteRT graphs for `CompiledModel`, converted with litert-torch: on a Galaxy S26, the Android sample in `android/` turns one photo into a 3D Gaussian splat of 65,536 Gaussians in 158 s with the default form, or of 262,144 Gaussians in 267 s with the first release's form in `262k/` (the first photo after start on a cool phone, LiteRT 2.3.0).
 
-[speridlabs/idol](https://huggingface.co/speridlabs/idol) turns one photo of an object into a 3D Gaussian splat of
-262,144 Gaussians that you can turn around, including the sides the picture never showed. This repo holds the
+[speridlabs/idol](https://huggingface.co/speridlabs/idol) turns one photo of an object into a 3D Gaussian splat (262,144 Gaussians at its
+defaults) that you can turn around, including the sides the picture never showed. This repo holds the
 author's denoiser with the [speridlabs/idol-turbo](https://huggingface.co/speridlabs/idol-turbo) LoRA (5 steps, no
 guidance), its octree and
 Gaussian decoders, the FLUX.2 VAE encoder, and the author's BiRefNet background remover. The DINOv3 image encoder is
@@ -28,50 +28,87 @@ in a separate, gated repo under its own licence:
 
 ![The photo, and the splat made from it turned to the front, the side and the back](assets/hero.png)
 
-*Left: [the photo](https://www.pexels.com/photo/plant-in-silver-flowerpot-20324601/) by Komet Flicker (Pexels License). Right: the splat the sample app made from it on a Galaxy S26, rendered from the front, the side and the back.*
+*Left: [the photo](https://www.pexels.com/photo/plant-in-silver-flowerpot-20324601/) by Komet Flicker (Pexels License). Right: the splat the sample app made from it on a Galaxy S26 with the default form (65,536 Gaussians), rendered from the front, the side and the back.*
 
-## Speed
+## Two forms
+
+The default form (the repo's root) is the fastest of the forms measured on a Galaxy S26: it makes a splat in 158 s against 267 s for the first release's form in `262k/`, while its renders agree less closely with the author's default output: 32.3–34.4 dB PSNR against 41.4–46.6 dB (higher is closer). Of the other forms tried, int8 weights did not run on the phone GPUs, a 512 × 512 input made even the author's own PyTorch pipeline miss the 25 dB bar (13.5–22.9 dB against its 1,024 × 1,024 output), the denoiser did not compile for the Galaxy S26's NPU on the phone, and the CPU alone was slower.
+
+| | default (the repo's root) | first release (`262k/`) |
+|---|---|---|
+| Gaussians per splat | 65,536 (2,048 anchors) | 262,144 (8,192 anchors, the author's default) |
+| GPU precision | FP16 | FP16 with fp32 accumulation |
+| Denoiser blocks 1–22 and Gaussian decoder graphs 1–6 | one shared graph per chain; the app copies each block's weights in before the call | a graph per block |
+| Galaxy S26: the first photo after start, cool phone (thermal status 0 at the start) | **158 s** | 267 s |
+| Galaxy S26: photos on a phone already warm | 329–351 s (4 photos of the armchair, thermal status 3) | 339–449 s (5 photos of 3 inputs, thermal status 2) |
+| Pixel 11 Pro: the first photo after start | 343 s | not measured |
+| 4-view render PSNR against the author's default on the Galaxy S26 (threshold 25 dB) | 32.3–34.4 dB | 41.4–46.6 dB |
 
-On a Galaxy S26, the sample app turns one photo into a splat of 262,144 Gaussians in 303 s, starting from a cool phone (the first photo after start). The phone throttles its GPU during that photo: the GPU clock cap fell from 1,300 to 422 MHz within its 5 minutes (thermal status 1 → 3). Photos taken one after another on a phone that was already warm (thermal status 2) took 339–449 s. The graphs do not all fit in the phone's memory at once, so the app closes graphs and compiles them again when the run comes back to them: about 115 compiles per photo, 70 s of the first one. Background removal (BiRefNet) runs on the CPU: 11.1 s of that time on the Galaxy S26. The
-other networks run on the GPU.
+Thermal status is Android's: 0 none, 1 light, 2 moderate, 3 severe. Both forms run with the same app: point `install_to_device.sh` at the repo's root or at `262k/`.
 
-Speed still being tuned (updates follow).
+The first release's card gives 303 s for the `262k/` form: that photo started on a warmer phone (thermal status 1, skin 38.5 °C, against 34.6 °C here). With this app the `262k/` form makes the same splat as with the first release's, byte for byte.
+
+The rest of this card describes the default form; the first release's card ([commit e7ee1226](https://huggingface.co/litert-community/Idol/blob/e7ee12262a3ed622fa44fb8df8ff4f335dd77064/README.md)) has the
+per-stage tables of the `262k/` form.
+
+## Speed
 
-| Galaxy S26 (SM-S942Q), the sample app, LiteRT 2.3.0; from a cool phone (GPU clock cap 1,300 MHz, thermal status 1 at the start) | first photo after start |
+On a Galaxy S26, the sample app turns one photo into a splat of 65,536 Gaussians in 158 s, starting from a cool phone (the first photo after start). The phone throttles its GPU during that photo: the GPU clock cap fell from 1,300 to 500 MHz (thermal status 0 → 2). After 12 minutes of back-to-back runs the phone was at thermal status 3; through the next 4 photos the GPU clock cap stayed at 222–282 MHz (one reading of 191 MHz), and they took 329–351 s. At start the app compiles 1 graph before the photo (2 s) and 19 more during it (20 s): the denoiser's 22 middle blocks share one graph and the Gaussian decoder's six middle graphs share another, so each is compiled once per photo. The first release's form in `262k/` (262,144 Gaussians, a graph per block, GPU FP16 with fp32 accumulation) took 267 s the same way, also from thermal status 0 (skin 34.6 °C at the start against 33.5 °C). Background removal (BiRefNet) runs on the CPU: 7.4 s of that time on the Galaxy S26. The
+other networks run on the GPU at FP16.
+
+| Galaxy S26 (SM-S942Q), the sample app, LiteRT 2.3.0; from a cool phone (GPU clock cap 1,300 MHz, thermal status 0 at the start) | first photo after start |
+|---|---|
+| **photo → .ply written** | **158 s** |
+| background removal (BiRefNet, CPU) | 7.4 s |
+| DINOv3 encoder | 10.6 s |
+| VAE encoder | 5.2 s |
+| denoiser, 5 steps | 119.8 s |
+| octree decoder | 9.1 s |
+| Gaussian decoder | 5.5 s |
+| PLY writer | 0.1 s |
+| graphs compiled at app start, before the first photo | 1 (2 s) |
+| graphs compiled during the photo | 19 (20 s) |
+| peak resident memory (VmHWM) | 5.42 GiB |
+| peak GPU memory (kgsl) | 5.73 GiB |
+
+| Galaxy S26 (SM-S942Q), the sample app, LiteRT 2.3.0; right after the cool photo's hold (thermal status 3 at the start) | first photo after start | photo 2 | photo 3 | photo 4 |
+|---|---|---|---|---|
+| **photo → .ply written** | **351 s** | **351 s** | **341 s** | **329 s** |
+| background removal (BiRefNet, CPU) | 14.9 s | 18.9 s | 15.2 s | 15.4 s |
+| DINOv3 encoder | 22.5 s | 23.6 s | 25.1 s | 20.6 s |
+| VAE encoder | 10.7 s | 8.1 s | 8.0 s | 7.5 s |
+| denoiser, 5 steps | 278.1 s | 280.3 s | 272.6 s | 265.4 s |
+| octree decoder | 14.3 s | 11.7 s | 12.1 s | 12.2 s |
+| Gaussian decoder | 9.2 s | 7.7 s | 7.6 s | 7.4 s |
+| PLY writer | 0.1 s | 0.0 s | 0.0 s | 0.0 s |
+| graphs compiled at app start, before the first photo | 3 (9 s) | — | — | — |
+| graphs compiled during the photo | 18 (29 s) | 18 (21 s) | 19 (25 s) | 18 (20 s) |
+| peak resident memory (VmHWM) | 5.30 GiB | 4.58 GiB | 5.10 GiB | 4.63 GiB |
+| peak GPU memory (kgsl) | 5.73 GiB | 5.12 GiB | 5.13 GiB | 5.04 GiB |
+
+| Pixel 11 Pro, the sample app, LiteRT 2.3.0; thermal status 0 at the start | first photo after start |
 |---|---|
-| **photo → .ply written** | **303 s** |
-| background removal (BiRefNet, CPU) | 11.1 s |
-| DINOv3 encoder | 6.3 s |
-| VAE encoder | 3.7 s |
-| denoiser, 5 steps | 228.4 s |
-| octree decoder | 16.2 s |
-| Gaussian decoder | 36.0 s |
+| **photo → .ply written** | **343 s** |
+| background removal (BiRefNet, CPU) | 17.4 s |
+| DINOv3 encoder | 8.5 s |
+| VAE encoder | 36.1 s |
+| denoiser, 5 steps | 242.3 s |
+| octree decoder | 12.9 s |
+| Gaussian decoder | 23.7 s |
 | PLY writer | 0.2 s |
-| graphs compiled during the photo | 115 (70 s) |
-| peak resident memory (VmHWM) | 4.71 GiB |
-| peak GPU memory (kgsl) | 5.60 GiB |
-
-| Galaxy S26 (SM-S942Q), the sample app, LiteRT 2.3.0; thermal status 2 (moderate) from the first photo | first photo after start | photo 2 | photo 3 |
-|---|---|---|---|
-| **photo → .ply written** | **339 s** | **416 s** | **414 s** |
-| background removal (BiRefNet, CPU) | 8.5 s | 17.7 s | 15.9 s |
-| DINOv3 encoder | 8.8 s | 31.0 s | 26.7 s |
-| VAE encoder | 5.4 s | 8.5 s | 7.3 s |
-| denoiser, 5 steps | 262.9 s | 306.9 s | 311.9 s |
-| octree decoder | 16.0 s | 15.1 s | 15.3 s |
-| Gaussian decoder | 35.6 s | 35.7 s | 36.2 s |
-| PLY writer | 0.2 s | 0.1 s | 0.1 s |
-| graphs compiled during the photo | 115 (78 s) | 114 (97 s) | 116 (93 s) |
-| peak resident memory (VmHWM) | 4.91 GiB | 4.16 GiB | 4.09 GiB |
-| peak GPU memory (kgsl) | 5.60 GiB | 5.33 GiB | 5.45 GiB |
+| graphs compiled at app start, before the first photo | 6 (60 s) |
+| graphs compiled during the photo | 15 (75 s) |
+| peak resident memory (VmHWM) | 5.96 GiB |
 
 ## How close it is to the author's model
 
 Each check runs the author's PyTorch pipeline (float32, CPU) and these graphs on the same input with the same random
 numbers, against thresholds fixed before the first run. The inputs are two images from the author's gallery
 (`nb-armchair`, `fig-axolotl`) and one photo (a succulent, Pexels). Per part, each graph gets the author's own input
-for that part and its output is compared with the author's (cosine; the threshold is 0.99). The whole pipeline is compared
-by rendering both splats from four sides with the same renderer (PSNR; the threshold is 25 dB).
+for that part and its output is compared with the author's (cosine; the threshold is 0.99); the octree and Gaussian
+decoders are compared with the author's model run at this form's 65,536 Gaussians (`--num-gaussians 65536`). The
+whole pipeline is compared with the author's default (262,144 Gaussians) by rendering both splats from four sides with
+the same renderer (PSNR; the threshold is 25 dB).
 
 ### Per-part agreement on the Mac (LiteRT GPU, Metal, fp16): cosine vs the author's model
 
@@ -79,40 +116,40 @@ by rendering both splats from four sides with the same renderer (PSNR; the thres
 |---|---|---|---|---|
 | DINOv3 encoder | feature1 | 0.99989 | 0.99994 | 0.99992 |
 | VAE encoder | latent tokens | 0.99995 | 0.99993 | 0.99996 |
-| Denoiser (5 steps) | latent velocity, worst step | 0.99931 | 0.99923 | 0.99931 |
-| Octree decoder (8 levels) | child probabilities, worst level | 0.99946 | 0.99937 | 0.99943 |
+| Denoiser (5 steps) | latent velocity, worst step | 0.99931 | 0.99924 | 0.99931 |
+| Octree decoder (8 levels) | child probabilities, worst level | 0.99948 | 0.99937 | 0.99944 |
 | Gaussian decoder | xyz | 1.00000 | 1.00000 | 1.00000 |
-|  | colour (f_dc) | 0.99970 | 0.99963 | 0.99976 |
-|  | scale | 0.99833 | 0.99837 | 0.99652 |
-|  | rotation | 0.99891 | 0.99915 | 0.99901 |
-|  | opacity | 0.99987 | 0.99871 | 0.99988 |
+|  | colour (f_dc) | 0.99980 | 0.99987 | 0.99937 |
+|  | scale | 0.99959 | 0.99925 | 0.99225 |
+|  | rotation | 0.99953 | 0.99919 | 0.99929 |
+|  | opacity | 0.99995 | 0.99969 | 0.99969 |
 
-### Per-part agreement on a Galaxy S26 (the app; GPU fp16 with fp32 accumulation, BiRefNet on the CPU): cosine vs the author's model
+### Per-part agreement on a Galaxy S26 (the app; GPU FP16, BiRefNet on the CPU): cosine vs the author's model
 
 | Part | Compared tensor | armchair |
 |---|---|---|
-| DINOv3 encoder | feature1 | 1.00000 |
-| VAE encoder | latent tokens | 0.99999 |
-| Denoiser (5 steps) | latent velocity, worst step | 0.99999 |
-| Octree decoder (8 levels) | child probabilities, worst level | 0.99994 |
+| DINOv3 encoder | feature1 | 0.99987 |
+| VAE encoder | latent tokens | 0.99991 |
+| Denoiser (5 steps) | latent velocity, worst step | 0.99924 |
+| Octree decoder (8 levels) | child probabilities, worst level | 0.99943 |
 | Gaussian decoder | xyz | 1.00000 |
-|  | colour (f_dc) | 0.99997 |
-|  | scale | 0.99990 |
-|  | rotation | 0.99986 |
-|  | opacity | 0.99993 |
+|  | colour (f_dc) | 0.99973 |
+|  | scale | 0.99939 |
+|  | rotation | 0.99938 |
+|  | opacity | 0.99996 |
 
-### Per-part agreement on a Pixel 11 Pro (the app; GPU fp16 with fp32 accumulation, BiRefNet on the CPU): cosine vs the author's model
+### Per-part agreement on a Pixel 11 Pro (the app; GPU FP16, BiRefNet on the CPU): cosine vs the author's model
 
 | Part | Compared tensor | armchair |
 |---|---|---|
-| DINOv3 encoder | feature1 | 0.99998 |
-| VAE encoder | latent tokens | 0.99999 |
-| Denoiser (5 steps) | latent velocity, worst step | 0.99989 |
-| Octree decoder (8 levels) | child probabilities, worst level | 0.99970 |
+| DINOv3 encoder | feature1 | 0.99989 |
+| VAE encoder | latent tokens | 0.99997 |
+| Denoiser (5 steps) | latent velocity, worst step | 0.99930 |
+| Octree decoder (8 levels) | child probabilities, worst level | 0.99945 |
 | Gaussian decoder | xyz | 1.00000 |
-|  | colour (f_dc) | 0.99989 |
-|  | scale | 0.99964 |
-|  | rotation | 0.99976 |
+|  | colour (f_dc) | 0.99988 |
+|  | scale | 0.99963 |
+|  | rotation | 0.99939 |
 |  | opacity | 0.99998 |
 
 ### Background removal
@@ -129,19 +166,20 @@ by rendering both splats from four sides with the same renderer (PSNR; the thres
 
 | Input | Where | PSNR, 4 views pooled (threshold: 25 dB) | per view: 0°, 90°, 180°, 270° |
 |---|---|---|---|
-| armchair | Mac, LiteRT GPU (Metal, fp16) | 35.51 dB | 34.86, 37.92, 34.15, 36.00 |
-| armchair | Galaxy S26, the app | 46.59 dB | 46.53, 45.59, 46.41, 48.25 |
-| axolotl | Mac, LiteRT GPU (Metal, fp16) | 32.21 dB | 33.85, 29.70, 34.58, 32.43 |
-| axolotl | Galaxy S26, the app | 42.34 dB | 40.72, 44.21, 40.59, 46.32 |
-| succulent | Mac, LiteRT GPU (Metal, fp16) | 36.38 dB | 35.41, 36.84, 36.14, 37.39 |
-| succulent | Galaxy S26, the app | 41.43 dB | 40.30, 41.09, 42.18, 42.51 |
+| armchair | Mac, LiteRT GPU (Metal, fp16) | 35.02 dB | 34.23, 36.99, 33.39, 36.48 |
+| armchair | Galaxy S26, the app | 34.41 dB | 34.22, 35.85, 33.69, 34.15 |
+| armchair | Pixel 11 Pro, the app | 33.58 dB | 32.42, 34.81, 31.90, 36.89 |
+| axolotl | Mac, LiteRT GPU (Metal, fp16) | 31.24 dB | 32.84, 28.79, 33.37, 31.52 |
+| axolotl | Galaxy S26, the app | 32.26 dB | 33.27, 30.07, 33.71, 33.04 |
+| succulent | Mac, LiteRT GPU (Metal, fp16) | 34.85 dB | 34.05, 34.89, 35.03, 35.58 |
+| succulent | Galaxy S26, the app | 34.26 dB | 33.47, 34.26, 34.41, 35.05 |
 
 ## Minimal usage
 
 ### Python: the whole pipeline
 
 ```bash
-hf download litert-community/Idol --local-dir Idol
+hf download litert-community/Idol --exclude "262k/*" --local-dir Idol   # the default form; drop --exclude for both
 cd Idol
 pip install -r python/requirements.txt
 python python/run_idol.py photo.jpg --out out --models .   # writes out/gaussian.ply
@@ -149,7 +187,7 @@ python python/run_idol.py photo.jpg --out out --models .   # writes out/gaussian
 
 `run_idol.py` downloads the DINOv3 repo (accept its licence on the Hub and `hf auth login` first) and runs every
 graph with `CompiledModel`, the GPU for all but BiRefNet. The steps between graphs are in `python/idol_runner.py` and
-`python/idol_glue.py`. On an M4 Max Mac (Metal GPU) one photo took 64 s and up to 22 GB of memory.
+`python/idol_glue.py`. On an M4 Max Mac (Metal GPU) one photo took 53 s and up to 6 GB of memory (the process footprint).
 
 ### Python: one graph with `CompiledModel` (background removal)
 
@@ -191,7 +229,7 @@ val alpha = outputs[0].readFloat() // [1024 * 1024] sigmoid map, 1 = the object
 for (b in inputs + outputs) b.close()
 bir.close()
 val gpu = CompiledModel.Options(Accelerator.GPU).apply {   // every other graph
-    gpuOptions = CompiledModel.GpuOptions(precision = CompiledModel.GpuOptions.Precision.FP16_WITH_FP32_ACCUM)
+    gpuOptions = CompiledModel.GpuOptions(precision = CompiledModel.GpuOptions.Precision.FP16)
 }
 ```
 
@@ -199,12 +237,13 @@ The Android sample app (photo or camera → progress per stage → a splat you c
 The graphs are too large for the APK; a script copies them into the app's files folder over adb:
 
 ```bash
-hf download litert-community/Idol --local-dir ~/Downloads/Idol
+hf download litert-community/Idol --exclude "262k/*" --local-dir ~/Downloads/Idol   # the default form; drop --exclude to get 262k/ too
 hf download litert-community/DINOv3-ViT-H16plus-Idol-LiteRT --local-dir ~/Downloads/DINOv3-ViT-H16plus-Idol-LiteRT
 cd ~/Downloads/Idol/android
 chmod +x gradlew scripts/install_to_device.sh        # the Hub does not keep the executable bit
 ./gradlew :app:installDebug
 IDOL_DIR=~/Downloads/Idol DINO_DIR=~/Downloads/DINOv3-ViT-H16plus-Idol-LiteRT ./scripts/install_to_device.sh
+# the first release's form instead (download without --exclude first): IDOL_DIR=~/Downloads/Idol/262k, same DINO_DIR
 ```
 
 Background removal (BiRefNet on the CPU) needs about 2,670 MB of free memory for one call. Before compiling it, the
@@ -219,13 +258,20 @@ the call.
 | Background removal (BiRefNet) | `bir.tflite` | CPU | 1 |
 | DINOv3 ViT-H+/16 image encoder | `dino_0..3.tflite` (DINOv3 repo) | GPU | 4 |
 | FLUX.2 VAE encoder | `vae_0..3.tflite`, `vae_attn.tflite`, `vae_4.tflite` | GPU | 6 |
-| Denoiser with the turbo LoRA | `den_ctx.tflite`, `den_in.tflite`, `den_joint_00..23.tflite` | GPU | 5 steps × 25, plus `den_ctx` once per LoRA phase |
+| Denoiser with the turbo LoRA | `den_ctx.tflite`, `den_in.tflite`, `den_joint_00.tflite`, `den_joint_mid.tflite` (blocks 1–22, weights from `den_joint_w.f32`), `den_joint_23.tflite` | GPU | 5 steps × 25, plus `den_ctx` once per LoRA phase |
 | Octree decoder | `oct.tflite` (5 signatures by row count) | GPU | 8 levels |
-| Gaussian decoder | `gs_0..7.tflite` | GPU | 8 |
+| Gaussian decoder | `gs_0.tflite`, `gs_mid.tflite` (graphs 1–6, weights from `gs_w.f32`), `gs_7.tflite` | GPU | 8 |
 | Octree sampling, Gaussian activation, PLY | host code | CPU | — |
 
-`manifest.json` lists every graph with its signatures, accelerator and precision. On Android the GPU graphs run with
-`FP16_WITH_FP32_ACCUM` precision.
+`262k/` holds the first release's form: `den_joint_01..22.tflite` and `gs_1..6.tflite` in place of the two shared
+graphs, the Gaussian decoder at 8,192 anchors, and its own `manifest.json` (GPU precision FP16 with fp32 accumulation);
+the files with the same name in both folders are the same bytes, except `gs_0.tflite`, `gs_7.tflite` and
+`manifest.json`.
+
+`manifest.json` lists every graph with its signatures, accelerator and precision; the GPU graphs run with `FP16`
+precision. `den_joint_mid` and `gs_mid` are one graph each for blocks of the same shape: before each call the app
+copies that block's weights from the side file into the graph's weight inputs (`shared_blocks`, `block_weights` and
+`first_weight_input` in `manifest.json`), so each is compiled once per photo.
 
 ## Notes
 
@@ -243,7 +289,7 @@ the call.
   Gaussian decoder's first block; the chunked form loses 0.6 %. Keep it if you re-export.
 - **Keep each GPU call short on the Galaxy S26.** In the feasibility run the VAE encoder ran as one graph, about 9 s
   per GPU call; the screen compositor stalled and the process was killed (signal 9) on its fifth call. Every stage here
-  is split into several graphs for that reason (on the Galaxy S26 every GPU call takes 0.04–1.83 s: 45 graphs, 49 signatures).
+  is split into several graphs for that reason (with GPU fp16 with fp32 accumulation, every GPU call of the first form took 0.04–1.83 s on the Galaxy S26 (45 graphs, 49 signatures); this form calls the same blocks, or the Gaussian decoder's at a quarter of the anchors, at FP16).
 - **The denoiser did not compile for the NPU on the phone.** Compiled on the Galaxy S26 for its NPU (Qualcomm HTP;
   LiteRT 2.3.0, QAIRT 2.50), one denoiser block did not finish in 15 minutes at the default optimization level (O3)
   and ran out of memory after 28 s at `optimization_level=1` (optimize for prepare); the HTP compiler warned it has

android/app/src/main/java/com/idol/BlockWeights.kt

@@ -0,0 +1,69 @@
+package com.idol
+
+import java.io.File
+import java.io.RandomAccessFile
+import java.nio.ByteOrder
+import java.nio.FloatBuffer
+import java.nio.channels.FileChannel
+
+/**
+ * The weights of the blocks one shared graph runs (the denoiser's middle joint blocks in `den_joint_mid`, the Gaussian
+ * decoder's middle graphs in `gs_mid`): a little-endian float32 side file, block after block, each block's weight
+ * inputs in input-name order (from args_<first_weight_input> on). The graph is compiled once per photo instead of once
+ * per block; before each call the block's weights are copied from the mapped file into the graph's input buffers.
+ */
+class BlockWeights(file: File, private val spec: GraphSpec) : AutoCloseable {
+    private val raf = RandomAccessFile(file, "r")
+    private val floats: FloatBuffer
+
+    /** For each weight input: its index in the signature's input list, its offset in one block (floats), its size. */
+    private val slots: List<Triple<Int, Int, Int>>
+    private val perBlock: Int
+
+    /** One array per distinct input size, reused slot after slot (a write copies it into the input buffer). */
+    private val scratch: Map<Int, FloatArray>
+
+    init {
+        try {
+            val sig = spec.signature()
+            val indexOf = sig.inputs.withIndex().associate { it.value.name to it.index }
+            val names = sig.inputs.map { it.name }.filter { trailing(it) >= spec.firstWeightInput }
+                .sortedBy { trailing(it) }
+            var offset = 0
+            slots = names.map { n ->
+                val i = indexOf.getValue(n)
+                val size = sig.inputs[i].size
+                Triple(i, offset, size).also { offset += size }
+            }
+            perBlock = offset
+            val expect = perBlock.toLong() * spec.sharedBlocks.size * 4
+            require(file.length() == expect) { "${file.name}: ${file.length()} bytes, expected $expect" }
+            floats = raf.channel.map(FileChannel.MapMode.READ_ONLY, 0, file.length())
+                .order(ByteOrder.LITTLE_ENDIAN).asFloatBuffer()
+            scratch = slots.map { it.third }.distinct().associateWith { FloatArray(it) }
+        } catch (e: Throwable) {
+            raf.close()
+            throw e
+        }
+    }
+
+    /** Writes block [block]'s weights into [g]'s inputs (tagged: a block already in the buffers is not copied again). */
+    fun write(g: Graph, block: Int) {
+        val pos = spec.sharedBlocks.indexOf(block)
+        require(pos >= 0) { "block $block is not one of ${spec.name}'s shared blocks" }
+        val base = pos.toLong() * perBlock
+        for ((k, slot) in slots.withIndex()) {
+            val (index, offset, size) = slot
+            val a = scratch.getValue(size)
+            floats.position((base + offset).toInt())
+            floats.get(a, 0, size)
+            g.write(index, a, "w${block}_$k")
+        }
+    }
+
+    override fun close() {
+        raf.close()
+    }
+
+    private fun trailing(name: String): Int = name.substringAfterLast('_').toIntOrNull() ?: -1
+}