{
  "experiment": "ComfyUI warm-cache dependency audit",
  "recorded_at": "2026-07-23",
  "environment": {
    "environment_label": "Movey test GPU",
    "comfyui_version": "0.28.2",
    "python_version": "3.13.8",
    "gpu": "NVIDIA GeForce RTX 3090"
  },
  "workflow": {
    "model": "Z-Image-Turbo",
    "lora": null,
    "text_encoder": "Qwen3 4B through ComfyUI Lumina2 integration",
    "vae": "ae.safetensors",
    "prompt": "A cute puppy",
    "width": 768,
    "height": 432,
    "steps": 9,
    "cfg": 1,
    "sampler": "res_multistep",
    "scheduler": "simple",
    "denoise": 1
  },
  "observations": [
    {
      "submission": "warm_run",
      "wall_time_seconds": 4.202,
      "cached": [
        "model loader",
        "model sampling patch",
        "text encoder loader",
        "prompt conditioning",
        "zeroed conditioning",
        "VAE loader"
      ],
      "executed": [
        "latent creation",
        "sampler",
        "VAE decode",
        "save image"
      ]
    },
    {
      "submission": "exact_repeat",
      "wall_time_seconds": 0.102,
      "cached": [
        "all computational nodes"
      ],
      "executed": [
        "no image computation"
      ]
    },
    {
      "submission": "save_prefix_only_changed",
      "wall_time_seconds": 0.102,
      "cached": [
        "all nodes through VAE decode"
      ],
      "executed": [
        "save image"
      ]
    }
  ],
  "caveat": "This records warm-cache behavior on one active server. It is not a cold-start benchmark or a general speed claim."
}