Load Face Swap Model (mtb)

This is the loader for MTB's face-swap toolchain. It reads a face-swap model off disk and outputs it for the node that actually performs the swap. It's the MTB equivalent of what ReActor and Roop do - and under the hood it's literally the same engine as those, because they all load the same model.

That's worth understanding before you go shopping between face-swap packs. The model this loads is InsightFace's inswapper (inswapper_128.onnx), and every mainstream open face swapper - ReActor, Roop, FaceFusion, and this MTB path - is that one model wearing a different front end. Switching packs changes the UI, not the identity quality ceiling and not the license position. So if you've used ReActor, you already know what results to expect here.

How it works

Face swapping is a two-model job. This node loads the swapper (inswapper), which transplants one face onto another. Separately you load a face analysis model (see MTB's Load Face Analysis Model - InsightFace's buffalo_l), which detects and embeds the faces so the swapper knows what it's working with. The apply node takes both plus your images.

The input and output

• faceswap_model - a dropdown populated from the model files you've placed in the face-swap models folder. In a fresh install this list is empty, because MTB doesn't ship the weights - you have to add inswapper_128.onnx yourself. Once it's on disk, it shows up here.

The single output is FACESWAP_MODEL, which wires into MTB's face-swap apply node alongside the loaded face analysis model.

How to install it

ComfyUI Manager: search MTB Nodes (comfy_mtb), install, restart. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/melMass/comfy_mtb, then restart. The face nodes lean on InsightFace and onnxruntime, which are optional dependencies - and historically the single most painful install in all of ComfyUI. If the face nodes don't load, MTB points you at http://127.0.0.1:8188/mtb to install the missing deps. Good news: InsightFace 1.0 (May 2026) finally dropped the C++ build requirement that caused years of "why won't this compile" grief, so a fresh install today is far less likely to send you hunting for Visual Studio build tools.

Common issues

Empty dropdown is the number-one confusion, and it's not a bug - there's just no model file yet. Download inswapper_128.onnx and drop it in the face-swap models folder, then restart or refresh; it'll appear.

Two things to set expectations. First, resolution: inswapper is frozen at 128×128 and officially unmaintained - the maintainers point anyone wanting higher resolution at their paid service. So the swapped face comes out low-res by modern standards. The standard fix is to run a face enhancer afterward (MTB's Load Face Enhance Model / GFPGAN), which cleans up and upscales the swapped face. Second, licensing: the InsightFace weights are non-commercial (the code is MIT, the models aren't), which quietly makes anything you build on this unsellable without a separate license. Fine for personal and research work; a real problem if you're shipping a product.