ClipVisionClamp

ClipVisionClamp is the CLIP_VISION version of Allor's pass-through clamps: feed it a CLIP_VISION object and it returns that exact object, unchanged. It's a no-op with a single input (clip_vision, type CLIP_VISION) and a single output of the same type. If you already understand ClipClamp, you understand this - same idea, different data type.

CLIP_VISION is the model object you get from a CLIPVisionLoader, the thing that encodes an image into the vision-encoder space used by IP-Adapter, style transfer, and image-prompt workflows. It's usually a one-to-one wire: loader → the node that consumes it. So why would you insert a node that does literally nothing?

Same three reasons as every clamp in this family:

• Splitting. If one loaded vision model feeds several consumers (say an IPAdapter plus a CLIP-vision-based analyzer), a clamp gives you a clean fan-out point instead of a tangle.
• Forcing order. The clamp creates a real graph dependency, so anything downstream of it executes after it. In a workflow where the vision model must be ready before a branch runs, that's a free scheduling guarantee.
• Readability. When you share a JSON workflow, an explicit named node communicates "this is the vision-model branch" better than an invisible wire into three places.

Practical notes

There's nothing to configure, so there's no real way to use it wrong. The only actual gotcha is a pack-level one: ClipVisionClamp ships in the Allor plugin, and installing the pack runs its requirements.txt, which installs rembg and onnx for the segmentation nodes even if you only want the clamps. On a fresh environment that's a heavier install than this node deserves. If that bugs you, this one node alone isn't worth the pack - it's a convenience for people already running Allor's image tools.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Nourepide/ComfyUI-Allor
cd ComfyUI-Allor
pip install -r requirements.txt

Then restart ComfyUI. Or just search "Allor" in ComfyUI Manager and click install. Two pack-specific things to know: first, Allor's git history was rebased to drop all images (repo shrank ~344x), which can throw off auto-updates - the docs at nourepide.github.io/ComfyUI-Allor-Doc have troubleshooting pages for that. Second, the vision model itself is not bundled; you still load it with ComfyUI's own CLIPVisionLoader.

Real talk

This is a "when you need it, you need it" node. Beginners can skip it entirely - wire CLIP_VISION straight through and move on. Where it earns its keep is the day you load one vision model and find yourself needing it in four places, or when you're debugging execution order in a big graph. Cheap, harmless, and it does exactly one thing: nothing.