PM Clip Loader

Most of the time you never think about the CLIP loader, because your checkpoint already contains a text encoder and "Load Checkpoint" hands it to you. PM Clip Loader exists for the other case: when the model you're running has a standalone text encoder that ships separately. That's the normal layout for a lot of Flux-style and diffusion-transformer pipelines, where you load the diffusion model with one node, the VAE with another, and the text encoder - a T5, a Qwen, a plain CLIP - with a third. This is that third node, wearing the PM Manager's nicer selection UI.

There's a real reason these encoders live on their own. The modern ones are huge - the encoder alone can be as heavy as the generation model - and shipping them separately lets you run the same encoder with different models and, more importantly, swap or upgrade it without touching anything else. So when you're assembling a workflow out of split files, this node is how the text side of your graph comes to life.

How it works

The schema is refreshingly empty: no inputs at all. The whole thing is the picker. A hidden widget - driven by the pack's frontend - holds the list of encoders you can choose, and when you pick one the node calls comfy.sd.load_clip on the file path. It looks in ComfyUI's clip folder (and handles the text_encoders layout some packs use), which is where those standalone T5/Qwen/CLIP files tend to live.

The output name logic is a small tell about how it's meant to be used: it strips the text_encoders/ or clip/ prefix and hands you the plain filename, so the model_name output gives you something clean to pass to a filename or metadata node.

Inputs and outputs

• clip - the loaded CLIP/text encoder. Wire it into a CLIP Text Encode node, same as you would the clip output of a checkpoint loader.
• model_name - a * string with the encoder's filename, mostly for bookkeeping.

That's it. If your workflow loads a CLIP and you don't remember wiring one in, this is likely what fed it.

Installing it

Standard PM Manager pack install - you're getting all eight nodes in one go:

cd ComfyUI/custom_nodes
git clone https://github.com/pixel-magician/ComfyUI-PM-Manager.git
# restart ComfyUI

Or use ComfyUI Manager and search for ComfyUI-PM-Manager. Its two dependencies (opencv-python, imageio-ffmpeg) install with it, and there are no model downloads - it reads whatever encoder files you've already put in models/clip.

Common issues

Because there are no explicit inputs, a "No CLIP model selected" error means the picker widget came up empty - which usually means the browser tab is holding stale state. Refresh and re-pick, and make sure your encoder file is actually in models/clip (or models/text_encoders). And once again: if the PM nodes are missing entirely, it's the pack's dependency on ComfyUI's newer extension API talking - update ComfyUI before you debug anything else.