{
  "id": "video-dit",
  "name": "Latent Video Diffusion Transformer",
  "description": "The shape every 2024-2025 video generator shares: a CAUSAL 3D VAE compresses the clip so frame t never sees frame t+1, the latent is patchified into space-time tokens, and a DiT stack denoises them under timestep conditioning. The causal padding is what lets the encoder run on a growing clip.",
  "icon": "🎞️",
  "category": "Generative",
  "components": [
    {
      "id": "in-video",
      "type": "input",
      "name": "Video clip (16 frames, 256x256)",
      "position": {
        "x": 240,
        "y": 60
      },
      "params": {
        "shape": [
          3,
          16,
          256,
          256
        ]
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "v1",
      "type": "causalConv3d",
      "name": "Causal VAE Down 128",
      "position": {
        "x": 240,
        "y": 160
      },
      "params": {
        "inChannels": 3,
        "outChannels": 128,
        "kernelSize": 3,
        "stride": 2,
        "temporalStride": 1,
        "padding": 1
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "v2",
      "type": "causalConv3d",
      "name": "Causal VAE Down 256",
      "position": {
        "x": 240,
        "y": 260
      },
      "params": {
        "inChannels": 128,
        "outChannels": 256,
        "kernelSize": 3,
        "stride": 2,
        "temporalStride": 2,
        "padding": 1
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "v3",
      "type": "causalConv3d",
      "name": "Causal VAE to latent 16",
      "position": {
        "x": 240,
        "y": 360
      },
      "params": {
        "inChannels": 256,
        "outChannels": 16,
        "kernelSize": 3,
        "stride": 2,
        "temporalStride": 2,
        "padding": 1
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "tub",
      "type": "tubeletEmbed",
      "name": "Patchify 1x2x2",
      "position": {
        "x": 240,
        "y": 460
      },
      "params": {
        "tubeletSize": [
          1,
          2,
          2
        ],
        "embedDim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "pos",
      "type": "positionalEncoding",
      "name": "Pos Encoding",
      "position": {
        "x": 240,
        "y": 560
      },
      "params": {
        "dModel": 1152,
        "maxLen": 1024
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "in-t",
      "type": "input",
      "name": "Diffusion timestep",
      "position": {
        "x": 620,
        "y": 60
      },
      "params": {
        "shape": [
          1
        ]
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "temb",
      "type": "timeEmbedding",
      "name": "Timestep Embed",
      "position": {
        "x": 620,
        "y": 160
      },
      "params": {
        "dim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "dit1",
      "type": "ditBlock",
      "name": "DiT Block 1 (AdaLN-Zero)",
      "position": {
        "x": 380,
        "y": 660
      },
      "params": {
        "hiddenDim": 1152,
        "numHeads": 16,
        "condDim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "dit2",
      "type": "ditBlock",
      "name": "DiT Block 2 (AdaLN-Zero)",
      "position": {
        "x": 380,
        "y": 760
      },
      "params": {
        "hiddenDim": 1152,
        "numHeads": 16,
        "condDim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "dit3",
      "type": "ditBlock",
      "name": "DiT Block 3 (AdaLN-Zero)",
      "position": {
        "x": 380,
        "y": 860
      },
      "params": {
        "hiddenDim": 1152,
        "numHeads": 16,
        "condDim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "dit4",
      "type": "ditBlock",
      "name": "DiT Block 4 (AdaLN-Zero)",
      "position": {
        "x": 380,
        "y": 960
      },
      "params": {
        "hiddenDim": 1152,
        "numHeads": 16,
        "condDim": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "norm",
      "type": "layerNorm",
      "name": "Final LayerNorm",
      "position": {
        "x": 380,
        "y": 1060
      },
      "params": {
        "normalizedShape": 1152
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "unpatch",
      "type": "linear",
      "name": "Unpatchify to latent",
      "position": {
        "x": 380,
        "y": 1160
      },
      "params": {
        "inFeatures": 1152,
        "outFeatures": 64
      },
      "inputs": [],
      "outputs": []
    },
    {
      "id": "out",
      "type": "output",
      "name": "predicted latent velocity",
      "position": {
        "x": 380,
        "y": 1260
      },
      "params": {},
      "inputs": [],
      "outputs": []
    }
  ],
  "connections": [
    {
      "id": "c1",
      "from": "in-video",
      "to": "v1",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c2",
      "from": "v1",
      "to": "v2",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c3",
      "from": "v2",
      "to": "v3",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c4",
      "from": "v3",
      "to": "tub",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c5",
      "from": "tub",
      "to": "pos",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c6",
      "from": "in-t",
      "to": "temb",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c7",
      "from": "pos",
      "to": "dit1",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c8",
      "from": "temb",
      "to": "dit1",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c9",
      "from": "dit1",
      "to": "dit2",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c10",
      "from": "temb",
      "to": "dit2",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c11",
      "from": "dit2",
      "to": "dit3",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c12",
      "from": "temb",
      "to": "dit3",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c13",
      "from": "dit3",
      "to": "dit4",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c14",
      "from": "temb",
      "to": "dit4",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c15",
      "from": "dit4",
      "to": "norm",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c16",
      "from": "norm",
      "to": "unpatch",
      "fromPort": "bottom",
      "toPort": "top"
    },
    {
      "id": "c17",
      "from": "unpatch",
      "to": "out",
      "fromPort": "bottom",
      "toPort": "top"
    }
  ]
}