Nodes/comfyui-sdnq-splited/SDNQ Model Loader
ComfyUI Node

SDNQ Model Loader

Load SDNQ quantized models with 50-75% VRAM savings. Supports FLUX, SD3, SDXL, video models, and more.

By ussoewwin·Created 8 months ago·Updated 8 months ago· 3
SDNQ Model Loader
    • model
    • vae
    ◄model_selectionFLUX.1-dev-qint8►
    ◄custom_model_path►
    ◄dtypebfloat16►
    ◄memory_modebalanced►
    ◄auto_downloadtrue►
    ◄matmul_precisionint8►
    ◄use_xformersfalse►
    ◄use_flash_attentionfalse►
    ◄use_sage_attentionfalse►
    ◄enable_vae_tilingfalse►
    Categoryloaders/SDNQ

    Inputs (10)

    NameTypeDefaultDescription
    model_selectionCOMBOFLUX.1-dev-qint8Select a pre-configured SDNQ model (auto-downloads from HuggingFace) or choose [Custom Path] to specify a local model directory
    custom_model_pathSTRINGLocal path to SDNQ model directory (only used when [Custom Path] is selected). Example: /path/to/model or C:\path\to\model
    dtypeCOMBObfloat16Model precision. bfloat16 recommended for FLUX (best quality/speed). float16 for older GPUs. float32 for CPU.
    memory_modeCOMBObalancedMemory management: 'gpu' = All on GPU (fastest, needs 24GB+ VRAM). 'balanced' = Model offloading (12-16GB VRAM). 'lowvram' = Sequential offloading (8GB VRAM, slowest).
    auto_downloadBOOLEANtrueAutomatically download model from HuggingFace if not found locally. Disable to only use local models.
    matmul_precisionoptCOMBOint8Precision for Triton quantized matmul. 'int8' is standard, 'fp8' for newer GPUs (Ada/Hopper), 'none' to disable optimization. Requires Linux/WSL.
    use_xformersoptBOOLEANfalseEnable xFormers memory-efficient attention for 10-45% speedup. Works with all memory modes (gpu/balanced/lowvram). Auto-fallback to SDPA if xformers not installed or incompatible. Requires: pip install xformers
    use_flash_attentionoptBOOLEANfalseEnable Flash Attention (FA) for faster inference and lower VRAM usage. Requires ComfyUI started with --use-flash-attention flag. Works with modern GPUs (Ampere+).
    use_sage_attentionoptBOOLEANfalseEnable Sage Attention (SA) for optimized attention computation. Requires ComfyUI started with --use-sage-attention flag. Provides better performance on supported GPUs.
    enable_vae_tilingoptBOOLEANfalseEnable VAE tiling for very large images (>1536px). Prevents out-of-memory errors on high resolutions. Minimal performance impact. Recommended for images >1536x1536.

    Outputs (2)

    NameTypeDescription
    modelMODEL—
    vaeVAE—