voice_audio_multi/model_specs/soulx_flashhead.json

{
  "schema_version": 1,
  "family": "soulx_flashhead",
  "display_name": "SoulX-FlashHead",
  "description": "Audio-driven streaming portrait video generation.",
  "category": "audio_generation",
  "status": "experimental",
  "tasks": [
    "sfx"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "auto"
  ],
  "runtime": {
    "tags": [
      "gguf",
      "stream"
    ]
  },
  "capabilities": {
    "sfx": [
      "prompt_generation"
    ]
  },
  "options": {
    "request": [
      {
        "name": "reference_image_file",
        "type": "path",
        "description": "Portrait image used to condition appearance.",
        "required": true
      },
      {
        "name": "height",
        "type": "int",
        "description": "Output height, divisible by the VAE spatial stride.",
        "min": 32,
        "default": 512,
        "required": false
      },
      {
        "name": "width",
        "type": "int",
        "description": "Output width, divisible by the VAE spatial stride.",
        "min": 32,
        "default": 512,
        "required": false
      },
      {
        "name": "num_inference_steps",
        "type": "int",
        "description": "Distilled denoising steps.",
        "min": 1,
        "default": 4,
        "required": false
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Generation seed.",
        "min": 0,
        "default": 42,
        "required": false
      },
      {
        "name": "color_correction_strength",
        "type": "float",
        "description": "Blend output color statistics toward the reference portrait.",
        "min": 0,
        "max": 1,
        "default": 1,
        "required": false
      },
      {
        "name": "audio_guidance_scale",
        "type": "float",
        "description": "Audio CFG scale for non-distilled pretrained checkpoints; distilled Lite/Pro do not use CFG.",
        "default": 3,
        "min": 0,
        "required": false
      }
    ],
    "session": [
      {
        "name": "exec_plan",
        "type": "string",
        "description": "JSON component execution plan. vae.impl is ggml (default) or cudnn. cudnn requires the Lite variant, CUDA backend, and cuDNN Conv3D build support. No artifacts or params are used.",
        "required": false,
        "default": "{}"
      }
    ],
    "load": []
  },
  "packages": [
    {
      "id": "soulx_flashhead_lite_q8_0",
      "display_name": "SoulX-FlashHead Lite Q8",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "SoulX-FlashHead-GGUF",
      "files": [
        "SoulX-FlashHead-Lite-Q8_0.gguf"
      ],
      "default": true
    },
    {
      "id": "soulx_flashhead_lite_bf16",
      "display_name": "SoulX-FlashHead Lite BF16",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "SoulX-FlashHead-GGUF",
      "files": [
        "SoulX-FlashHead-Lite-BF16.gguf"
      ],
      "default": false
    }
  ],
  "dependencies": [],
  "ui": {
    "recommended_package": "soulx_flashhead_lite_q8_0",
    "docs": [
      "docs/models/soulx_flashhead.md"
    ],
    "tags": [
      "GGUF"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json"
      },
      "tensors": {
        "dit": {
          "source": "weights:",
          "prefix": "dit"
        },
        "vae": {
          "source": "weights:",
          "prefix": "vae"
        },
        "audio": {
          "source": "weights:",
          "prefix": "audio"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json"
      },
      "tensors": {
        "dit": "model:dit.safetensors",
        "vae": "model:vae.safetensors",
        "audio": "model:audio.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/SoulX-FlashHead-GGUF",
      "revision": "main",
      "gated": false
    }
  }
}