DualCLIPLoader_motorway_edition

Half of why SDXL and Flux understand prompts as well as they do is that they don't use one text encoder - they use two, and their outputs get fused. DualCLIPLoader is the ComfyUI node that loads that pair in one shot, and this is its Motorway-ed twin from agilly1989/ComfyUI_agilly1989_motorway.

In plain ComfyUI you'd feed it clip_name1, clip_name2, and a type, and it hands you a CLIP object. On the Motorway the only difference is where the result goes: instead of a CLIP output socket, it writes the loaded CLIP to a key on the pipe.

The recipe you're actually choosing

The type dropdown is the whole ballgame, because it decides how the two files are interpreted:

• sdxl - clip-L + clip-G (OpenCLIP bigG). The two SDXL encoders, fused.
• flux - clip-L + T5-XXL. That T5 is why Flux understands natural language so well; it's a real language model, not an image-text alignment model.
• sd3 - pick two of clip-L / clip-G / T5 for the SD3 family.
• hunyuan_video - HunyuanVideo's text encoder plus its vision encoder.

The clip_name1 and clip_name2 dropdowns are populated from ComfyUI's models/text_encoders folder, so the files have to live there. device (default default, or cpu) forces the encoder onto the CPU - occasionally the escape hatch when a big T5 won't fit on your card, since the encoder is its own memory budget on top of the diffusion model.

The Motorway shape

Everything except the widgets becomes a key on this clone:

• OUTPUT_CLIP_key (default CLIP) - where the loaded CLIP lands on the bus. Downstream, a motorway-ed CLIPTextEncode reads it via its INPUT_clip_key.

The node has no INPUT_* keys at all, because it reads nothing from the pipe - it's a pure writer, like a loader should be. The optional MOTORWAY 🚌💨 input is how it joins an existing flow; leave it unplugged and it just writes to the bus it creates on the fly.

Install and caveats

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart ComfyUI. No extra Python deps. The model files are the real "dependency": for flux you need the T5-XXL encoder and a clip-L, for sdxl the clip-L + clip-G pair, all sitting in models/text_encoders. ComfyUI Manager will also install the pack if you search its title.

The honest gotcha, same as every node in this pack: the latest release (1.1.7) is stamped "HOTFIX - REMOVED MONKEYPATCH AND GENERATED NODES" in its own pyproject.toml, and the generated clone nodes are commented out of __init__.py. The Motorway core works; these motorway-ed clones may not show up in your menu until the author re-enables them. It's a beta pack, and the author is upfront that things break.

If you do get it running, the one thing that bites: pick the type that matches your model, or the encoders load and the CLIP encodes garbage. Wrong type with a flux pair is a classic "why is my text mushy" post.