h4 - Load Face Model

If you've ever built a workflow that keeps a consistent character across sessions, you know the pain: every new run you re-download the reference, re-crop it, re-feed it. h4_Live's FaceForge suite tries to solve that by turning a face into a saved file - and this node is the reload half of that trick. Give it a saved .safetensors face model and it hands you back the face identity plus its name, ready to wire into whatever FaceForge does with it.

Where it fits

FaceForge is h4_Live's face-manipulation mini-ecosystem, and the flow is a loop of three nodes: Build Face Model extracts an embedding from a face image, Save Face Model writes that embedding to disk, and this one loads it back. So the intended lifecycle is: build a face once, save it, and reload it from any workflow afterward without touching the original reference photo again. It's the same "identity as a reusable asset" idea you see in the character-consistency corner of the ecosystem - just packaged as a custom node family instead of a LoRA.

How it works

It reads from the face-models folder that the FaceForge suite uses, lists whatever .safetensors files are there, and lets you pick one via the face_model dropdown. Loading hands you two outputs:

• FACE_MODEL - the face embedding object, ready to feed the rest of the FaceForge pipeline.
• FACE_MODEL_NAME - the filename as a string, handy for naming saves, logging, or just knowing which face you're looking at.

That's the whole node. It's deliberately dumb: load, name, done.

The catch: you need the other half

The dropdown only has choices if you've actually saved face models first. Out of the box it shows just none, and that's not a bug - it's the node telling you there's nothing to load yet. If that's your situation, run the Save Face Model flow at least once before you expect this one to do anything.

Also worth knowing: the whole FaceForge suite is the most obscure corner of an already obscure pack. The author's README covers it in a couple of lines, there's no separate model download (the face embeddings are trained or extracted locally), and you should not expect a community of people swapping pre-built face files. If you're serious about character consistency you might get more mileage from a proper identity/LoRA workflow - but if you're already inside h4_Live and just want your built faces to survive a restart, this is the piece that makes the loop close.

Installing it

Same pack, same install:

cd ComfyUI/custom_nodes
git clone https://github.com/m3rr/h4_Live
# restart ComfyUI

or ComfyUI Manager → search "h4_Live." No extra Python dependencies, no model downloads - the only "model files" are the face files you create yourself.