IPAdapter Model Loader

This is the plain, manual way to load an IP-Adapter model - pick a file from a dropdown, get an IPADAPTER out, done. It's the older, lower-level sibling of the Unified Loader, and it exists for one reason the Unified Loader can't cover: it works with any filename. If you've got a community adapter, a renamed file, or something that doesn't fit the strict naming the Unified Loader demands, this is the node that loads it.

What it does and doesn't do

The trade is simple. The Unified Loader is convenient but rigid - it only recognizes files named exactly as the README says, and it auto-pairs the matching CLIP vision encoder. The Model Loader is the opposite: total freedom on the filename, but it loads only the adapter. It does not load CLIP vision for you. That's the part people miss.

So a working graph with this node has an extra piece. You load the adapter here, you separately load a CLIP vision model with ComfyUI's core Load CLIP Vision node, and you feed both into an apply node like IPAdapter Advanced - the adapter into its ipadapter input, the CLIP vision into its clip_vision input. Forget the CLIP vision half and the apply node has nothing to encode your reference with.

Inputs and outputs

There's exactly one input: ipadapter_file, a dropdown populated from whatever's sitting in ComfyUI/models/ipadapter. If that dropdown is empty, it's not broken - it means the folder is empty (or doesn't exist). Drop a .safetensors or .bin adapter in there, refresh, and it'll show up. The single output is IPADAPTER, the loaded bundle.

That's the whole node. It's deliberately dumb, and that's the appeal.

When to reach for it

• Community and non-standard models. The ecosystem has baked adapters the Unified Loader doesn't know about - the Ostris composition adapters, for instance. Load those here.
• You renamed something. Downloaded a file with a weird name and don't want to rename it to match the README? This loader doesn't care.
• You're debugging. When the Unified Loader throws "IPAdapter model not found," swapping in the Model Loader is a quick way to confirm the file itself is fine and the problem was the naming convention.

The flip side: because you're wiring CLIP vision manually, you own the pairing. SD1.5 adapters need the ViT-H encoder; SDXL adapters need ViT-bigG. Feed the wrong one and you'll get an error or garbage output - the Unified Loader normally protects you from this, and here you're on your own.

Installing the pack

Through ComfyUI Manager: open Custom Nodes Manager, search "IPAdapter plus", install, restart. Or from a terminal: cd ComfyUI/custom_nodes && git clone https://github.com/cubiq/ComfyUI_IPAdapter_plus, then restart ComfyUI. Put your adapter files in ComfyUI/models/ipadapter (create the folder if it's missing) and your CLIP vision encoders in ComfyUI/models/clip_vision. If a node in the pack refuses to load, update ComfyUI first - IP-Adapter tracks the latest core release closely.

For most people starting out, the Unified Loader is the better default because it removes the CLIP-vision-pairing footgun entirely. Graduate to the Model Loader when you specifically need a file the Unified Loader won't touch.