Quadruple CLIP Selector

Some modern checkpoints use more than one CLIP text encoder. Dual-encoder models became the norm with SDXL; the triple and quadruple arrangements followed with later architectures that lean on a couple of encoders plus a few distilled variants for different jobs. Managing that in a reusable workflow is a wiring nightmare if every CLIP has its own loader. IPT-QuadrupleClipSelector is the pack's way of handling the four-encoder case: four dropdowns, each picking a CLIP file, and one clip reference output that carries the whole selection as a single value.

It's the CLIP analogue of the pack's selector family - Checkpoint Selector, Lora Selector, Dual CLIP Selector, Triple CLIP Selector - all of which exist so your model choices are data, not a pile of per-run loaders. The clip reference output feeds the same image-info machinery: store it in image_info, save it to metadata, restore it later.

How it works

You pick clip_name1 through clip_name4 from dropdowns populated from your CLIP folder, and the node bundles them into one clip reference that downstream loaders resolve into actual runtime CLIP tensors when needed. There are four sha256_* optional inputs, cached hashes that back the right-click View Model Info... action - if the Civitai lookup can't resolve by filename, the cached SHA256 is the fallback key. Each clip_name also comes out as its own output, so you can echo the chosen names into filenames or captions.

Inputs and outputs

• clip_name1–clip_name4 (required, COMBO) - the four CLIP selections.
• clip (output, IPT-Clip) - the bundled reference.
• clip_name1–clip_name4 (outputs) - the individual selections as plain values.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/kinorax/comfyui-info-prompt-toolkit.git
cd comfyui-info-prompt-toolkit
pip install -r requirements.txt

Restart, or ComfyUI Manager → "ComfyUI-Info-Prompt-Toolkit". ComfyUI 0.17.0+. No models.

Common issues

If you don't have a four-encoder model, you don't need this node - the Dual and Triple selectors cover the common cases, and using this with fewer CLIPs means unused slots doing nothing but adding noise. Also remember the dropdown lists are built from your CLIP folder at node-construction time; a newly downloaded CLIP won't appear until the node refreshes. The sha256_* fields are socketless convenience inputs, so they don't need wiring unless you're chasing a View Model Info lookup that's coming up empty. For a quadruple-encoder workflow you're rebuilding anyway, this collapses four loaders into one value and earns its place.