Nodes/ComfyUI Ino Nodes/Ino Load VAE Model
ComfyUI Node

Ino Load VAE Model

Load a VAE from a path string when the file isn't a constant

By nobandegani·Created about a year ago·Updated 2 months ago· 6
Ino Load VAE Model
    • success
    • message
    • vae
    enabledtrue
    model_path

    The VAE is the quiet workhorse of a ComfyUI workflow - it's what turns latent space back into pixels at the end, and a bad VAE load means mush even when your sampler was perfect. Stock ComfyUI loads VAEs from a dropdown; Ino Load VAE Model loads one from a path string instead. Same reason you'd ever want that: the filename isn't known until runtime, whether it just got downloaded, or you're swapping VAEs per job, or you're building a fully config-driven pipeline.

    It's part of ComfyUI Ino Nodes (nobandegani/ComfyUI-InoNodes), the 125+-node V3-schema pack, and belongs to its InoModelHelper category - the family of loaders (UNET, CLIP, ControlNet, VAE, LoRA) that all take enabled + a path string and return success / message plus the loaded object.

    How it works

    It calls ComfyUI's own VAELoader under the hood, so there's no custom magic - and no custom magic is exactly the point. Because the path is a string input, another node can supply it. The natural upstream here is the pack's model-download nodes (HTTP, S3, HuggingFace, Civitai), which fetch a VAE file into models/vae and return its path. Download → load → decode, all as data flowing through the graph.

    Inputs are just two:

    • enabled - the pack-wide toggle; off means skip and report success: False.
    • model_path - path to the VAE file.

    Three outputs: success (boolean), message (string explaining failures), and vae (VAE type) - which wires straight into a VAE Decode node.

    When you'd reach for it

    The stock VAE Loader dropdown is genuinely fine for a fixed setup. You want this node when the workflow is doing any of the following: downloading models at runtime (a very Ino-flavored pattern), iterating over checkpoints with paired VAEs, or running the same graph on different machines where paths differ. It also composes with the pack's Ino Load Sampler Models, which bundles UNET + CLIP + VAE + LoRAs into one config-driven mega-node - this single loader is the same idea scaled down.

    Common issues

    • Bad path → success: False. The node won't crash the run; it reports failure in message and returns no VAE. In automation, gate your decode on success.
    • Wrong VAE for the checkpoint. This loader can't save you from loading an SD1.5 VAE into an SDXL graph - that's a workflow error, not a node error, and it usually shows up as weird colors or noise instead of a clean failure.
    • enabled off. Returns success: False, "not enabled" - expected behavior, not a bug.

    Installing it

    ComfyUI Ino Nodes installs like any custom node pack:

    • ComfyUI Manager (easiest): search for "ComfyUI Ino Nodes", install, restart.
    • Manual:
      cd ComfyUI/custom_nodes
      git clone https://github.com/nobandegani/ComfyUI-InoNodes.git
      cd ComfyUI-InoNodes
      pip install -r requirements.txt
      Then restart ComfyUI.

    Requires inopyutils and a current ComfyUI. It's a quiet, niche pack - the page's zero impressions say so - so the README is your manual. But if you've ever needed a VAE loaded from a path instead of a pick-list, this is the node.

    CategoryInoModelHelper

    Inputs (2)

    NameTypeDefaultDescription
    enabledBOOLEANtrue
    model_pathSTRING

    Outputs (3)

    NameTypeDescription
    successBOOLEAN
    messageSTRING
    vaeVAE