CLIP Vision Name

CLIP Vision is the model that looks at images. It's what powers IPAdapter-style image conditioning, InstantID and PuLID identity transfer, and any node that needs to understand "what is in this picture" before generation happens. CLIPVisionName is the picker for it: a dropdown of every vision model in your models/clip_vision folder, and the chosen filename out as a string.

It's one of the lhyNodes "name" family - the exact same value-node pattern as CheckpointName, CLIPName and ControlNetName in this pack - and like all of them its job is to be a single source of truth. If your workflow loads a CLIP Vision model for IPAdapter, and you switch base models a lot, having one dropdown that decides the vision model means you never run an SDXL-appropriate vision model against a Flux identity node by accident.

How it works

The dropdown enumerates folder_paths.get_filename_list("clip_vision") - same list CLIPVisionLoader shows. The selected name passes through untouched as the pack's "any" type, so it connects to any string socket. The intended wiring is the standard trick: right-click the loader (or the IPAdapter/InstantID node's clip_vision input) → Convert to input, and connect this node. One selection point, all consumers follow.

The sleeper use is metadata: wire the name into a filename prefix or a text node, and every image you save records which vision model conditioned it. When you're comparing identity-transfer outputs a week later, "why does this one drift?" is answered by a filename.

Inputs and outputs

• clip_name - dropdown of your CLIP Vision models (the field is named clip_name here rather than clip_vision_name; same thing).

One output: clip_name (*), the filename string.

Installing it

Part of ComfyUI-lhyNodes:

• ComfyUI Manager → search lhyNodes → Install, restart.
• Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/lihaoyun6/ComfyUI-lhyNodes.git
python -m pip install -r ComfyUI-lhyNodes/requirements.txt

Restart ComfyUI.

Where people get burned

The standard name-node trap applies with extra teeth: it returns a name, not a model. If you delete the actual CLIPVisionLoader and wire this straight into an IPAdapter node, the node gets a string where it expects a model and the run fails. And keep in mind the dropdown only lists what's in clip_vision - if your identity workflow uses a vision model bundled inside a different folder, it won't show up. As far as gotchas go, that's a mild one for a node whose entire personality is "pick one thing, output its name."