ClipVisionOutputClamp

ClipVisionOutputClamp is the pass-through node for the CLIP_VISION_OUTPUT data type - the actual encoded result a vision encoder produces. Same design as the rest of Allor's clamp family: one input in, the identical object out, no processing, no parameters, nothing to set. You'd be forgiven for asking what the point is, and the honest answer is "graph hygiene."

CLIP_VISION_OUTPUT is what you get out of a CLIPVisionEncode node: the image has already been pushed through the vision encoder, so it's not a model anymore, it's a result. In a typical workflow it feeds IP-Adapter or an image-prompt encoder and then you're done. But in the ones that get complicated - multi-pass style transfer, several adapters consuming the same encoding, conditional branches - you'll end up wanting that same output in more than one place.

That's where the clamp earns its keep:

• Fan-out without the knot. Run one encoding through ClipVisionOutputClamp and wire its single output to every consumer. The graph stays legible, and anyone who opens your saved workflow can see exactly where the vision-encoding branch starts.
• Forced execution order. Because the clamp is a real node, anything downstream of it is guaranteed to be scheduled after the encoding. That matters more than it sounds in graphs where lazy execution order is the source of weird nondeterminism.
• A named checkpoint in the pipeline. If you're iterating and want a stable "this is where the encoding lives" anchor you can inspect, a pass-through is a handy place to grab it.

Installing it

It ships inside the Allor plugin, so install that:

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

Restart ComfyUI (or search "Allor" in ComfyUI Manager). Two pack-level notes worth remembering: Allor's repo was rebased to strip images from git history, which can break auto-updates - the docs at nourepide.github.io/ComfyUI-Allor-Doc have a dedicated troubleshooting section for it. And the requirements.txt installs rembg plus onnx for the pack's segmentation nodes; you don't need them for this clamp, but they come along anyway.

Real talk

There's a running theme with Allor's clamps: they're all no-ops, and they're all optional. You won't need ClipVisionOutputClamp in your first hundred workflows. The day you load one vision model, encode one image, and find yourself wiring that result into four different adapters while trying to read your own canvas, this is the node that makes it tidy. Until then, skip it - it'll be here when the graph gets ugly.