SP_Name_ClipVision

A CLIP Vision model is the piece that reads a reference image rather than text - it's what IP-Adapter and similar style/face-transfer setups use to turn "this picture" into something the diffusion model can condition on, the same way a CLIP text encoder turns a prompt into conditioning. SP_Name_ClipVision doesn't load one of those; it picks a filename from your CLIP Vision folder and outputs it as plain text.

Why you'd want just the name

IP-Adapter workflows are one of the more common places where people swap components mid-experiment - different CLIP Vision encoders behave differently depending on which IP-Adapter checkpoint they're paired with, and getting the pairing wrong is a well-known source of bad results. If you're running comparisons across a few different vision-encoder files, this node lets you tag your output batch with which one was actually used, or drive a switch keyed on that choice, without the overhead of loading the encoder just to read its own name back.

Inputs and outputs that matter

• name - a dropdown populated from your CLIP Vision models folder.
• Output: STRING - the selected filename.

How to install it

Through ComfyUI Manager: search "SP-Nodes," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/bananasss00/ComfyUI-SP-Nodes

Restart ComfyUI once it's cloned. No extra dependency for this node itself.

Common issues & troubleshooting

Dropdown is empty. CLIP Vision models live in their own folder separate from checkpoints and regular CLIP files - if you've only ever installed a checkpoint and a base CLIP, you likely don't have anything here yet. Download a CLIP Vision model matching whatever IP-Adapter setup you're using, and restart ComfyUI so it picks up the new folder contents.

Right file exists but doesn't show up. Same startup-scan issue as the rest of this pack's SP_Name_* nodes - a file added while ComfyUI is already running won't appear until restart.

Getting this confused with the CLIP Vision loader. This node never runs the vision encoder and never touches your reference image - it only outputs a filename string. If your workflow needs the actual encoded output for an IP-Adapter chain, you still need a real CLIP Vision Loader node feeding into the IP-Adapter nodes directly; this one is for labeling and logic, not for the pipeline itself.

Not sure which CLIP Vision file your IP-Adapter setup actually needs. That's determined by the IP-Adapter checkpoint, not by this node - most IP-Adapter model cards state explicitly which CLIP Vision encoder they were trained against, and mismatching the two is a more common source of weak or broken style transfer than any single setting inside the IP-Adapter node itself. Check that pairing before assuming a bad result is a prompt or weight problem.