ClipVision Select (Uni)
Name your vision encoder, let the loader load it
- clip_vision_name
ClipVision Select (Uni) is a dropdown over your CLIP vision models that outputs the chosen filename as a string, with a none option for the Unified Loader. It's the vision-encoder member of the PG selector family, and it exists for a specific reason: CLIP vision models are the quiet workhorses behind IP-Adapter, and this node is how you hand the right one to the loader without hunting through widget menus.
A CLIP vision encoder is what turns an image into the semantic embedding that IP-Adapter uses to condition on a reference photo. When you're doing character consistency or style transfer from a reference image, the pairing of vision encoder + IP-Adapter is half the recipe - grab the wrong encoder and your style reference turns into mush. Which encoder you need depends on the IP-Adapter variant you're running (ViT-H for the standard line, ViT-bigG for the Plus v2 line), which is why keeping the choice explicit pays off.
How it works
At import it scans folder_paths for the clip-vision list, trying clip_vision, clip_vision_models, vision_models, and clip_vision_encoder keys so it survives folder renames by other packs. process() returns the selected name on the clip_vision_name output. As with all PG selectors: no loading, just a filename string for the Unified Loader to consume.
The input and output
clip_vision_name- dropdown over yourclip_visionfiles, defaultnone.- Output:
clip_vision_name- the filename as a string.
Install
Ships with PG Nodes:
ComfyUI Manager → search "PG Nodes" → Install → restart
Manual: git clone https://github.com/GizmoR13/PG-Nodes into custom_nodes (folder named PG-nodes), restart. Under PG/Select/UniLoader. No dependencies.
Common issues
- Wrong style transfer results. Before blaming the node, verify your vision encoder matches your IP-Adapter variant. ViT-bigG with a ViT-H IP-Adapter is a classic mismatch.
- File missing from the list. Restart ComfyUI; the dropdown materializes at import.
- Output won't reach an IPADAPTER input. It's a name, not the model. It feeds the loader's clip-vision slot.
The Uni selectors are the "extra" tier of the pack - the plain variant (ClipVision Select) covers non-Unified loaders, the Uni one feeds the combined loader. If you run IP-Adapter workflows through the Unified Loader, this is the matching knob; otherwise it's skippable. Useful, honest, and exactly as small as it looks.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| clip_vision_name | COMBO | none | 1 options: none |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| clip_vision_name | none | — |