ComfyUI Extension: ComfyUI Prepack

Authored by S4MUEL-404

Created

Updated

0 stars

A small, practical bundle of ComfyUI nodes that streamlines common workflows.

Custom Nodes (0)

    README

    💀Prepack – ComfyUI custom nodes (v1.0.0)

    A small, practical bundle of ComfyUI nodes that streamlines common workflows:

    • Unified loader for UNet + Dual-CLIP
    • Apply up to 3 LoRA adapters in one node and emit a prompt string
    • KSampler + VAE decode in one step
    • Pack/Unpack a pipeline into a single object
    • Frontend-only "virtual" Set/Get helpers for wiring arbitrary values

    This package is designed to minimize graph clutter while staying compatible with stock ComfyUI behaviors.

    Installation

    1. Copy the ComfyUI-Prepack folder into ComfyUI/custom_nodes/.
    2. Start or restart ComfyUI.
    3. You should see a new category named 💀Prepack in the node menu.

    Frontend extension (for the virtual Set/Get nodes) is automatically loaded from ./js via WEB_DIRECTORY.

    Nodes

    1) 💀Prepack UnetModel

    Load a diffusion UNet and a dual-CLIP pair in one step.

    • Inputs
      • unet_name: checkpoint from diffusion_models
      • weight_dtype: default | fp8_e4m3fn | fp8_e4m3fn_fast | fp8_e5m2
      • clip_name1, clip_name2: checkpoints from text_encoders
      • type: sdxl | sd3 | flux | hunyuan_video (falls back to SD if unavailable)
      • device (optional): default | cpu
    • Outputs: MODEL, CLIP
    • Notes: Float8 dtypes are applied only when your PyTorch build supports them.

    2) 💀Prepack Loras

    Apply up to three LoRA adapters to both model and CLIP at once and return a formatted prompt snippet.

    • Inputs
      • model, clip
      • lora_name_1 (required), lora_name_2 (optional), lora_name_3 (optional)
      • strength_model_*, strength_clip_* (typical range [-2.0, 2.0], 0 disables)
    • Outputs: MODEL, CLIP, STRING (lora_text)
    • The lora_text is a comma-separated list like <lora:YourLoRA:1.0> which you can add to prompts.

    3) 💀Prepack Ksampler

    Run KSampler and immediately decode the result with VAE.

    • Inputs: model, vae, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise
    • Output: IMAGE
    • Behavior: Handles empty latent channels and provides lightweight shape logs.

    4) 💀Prepack SetPipe / 💀Prepack GetPipe

    Pack a mini pipeline into a single PIPE object and unpack later.

    • SetPipe inputs: model, clip, lora_text, vae → output PIPE
    • GetPipe input: PIPE → outputs model, clip, lora_text, vae
    • Use case: Save/restore a working bundle for later branches.

    5) 💀Prepack Set Any / 💀Prepack Get Any (frontend-only)

    Lightweight virtual nodes that help you pass arbitrary values around the graph without polluting the prompt.

    • Location: Category 💀Prepack. Search terms: "prepack set any" / "prepack get any"
    • Behavior
      • Set Any: one text widget named "Constant"; one input and one output with mirrored type
      • Get Any: a combo widget that lists all Set Any names in the current graph
      • Types are propagated so connections remain valid
      • Widgets are persisted in the workflow (serialize enabled)
    • Note: These nodes are virtual and not serialized into the execution prompt.

    Tips & Troubleshooting

    • Missing nodes in the menu after install: restart ComfyUI and hard-refresh the browser (Ctrl/Cmd+Shift+R).
    • If float8 dtypes are unavailable, the loader silently falls back to default dtype.
    • If hunyuan_video CLIP type is not supported by your ComfyUI build, it falls back to SD.
    • Set/Get Any names persist in the workflow file; on load we re-sync titles and types automatically.

    Version

    • 1.0.0 – Initial public release.

    License

    Same license as this repository (see repo settings). If none is specified, treat as "All rights reserved" by default.