Load DualCLIP (Quantized, Simple)

The "Simple" suffix here means exactly what it means on the single-encoder loader: this is Load DualCLIP (Quantized) with the quant_format and kernel_backend dropdowns removed. Under the hood it calls the full loader with format set to auto and backend set to pytorch, then auto-detects whatever quant layout each of your two encoder files uses. You pick the files and the architecture, it does the rest.

This is the node you grab when you have a pair of quantized text encoders from silveroxides' HuggingFace - say a quantized CLIP-L and T5 for Flux - and you don't care about the format details, you just want both loaded into one CLIP object so your prompt encoder has something to work with. The full version's extra controls only earn their keep when you're mixing formats deliberately or troubleshooting a file that won't load.

The inputs that matter

• text_encoder1 and text_encoder2 - the two files, in the order your type expects. The second picker also lists checkpoints, and for the ltxv type it resolves from there.
• type - twelve presets (sdxl, sd3, flux, hunyuan_video, hidream, hunyuan_image, hunyuan_video_15, kandinsky5, kandinsky5_image, ltxv, newbie, ace). This one isn't auto-detected, so get it right - it decides which encoder combination ComfyUI builds.
• disable_dynamic and low_memory - the shared loading toggles (fast low-impact loading vs. ComfyUI's default).

Output is a single CLIP into CLIP Text Encode.

Installing it

Standard pack install: ComfyUI Manager → search "ComfyUI-QuantOps", or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-QuantOps

Restart, and let it pull in unifiedefficientloader>=0.5.2.

The catch

Two things worth knowing. "Auto" only auto-detects the pack's own recognized layouts (int8, tensorwise int8, fp8 variants, mxfp8, nvfp4) - a file that isn't one of those won't load just because this node is called "Simple." And the whole pack is deprecated: int8 ConvRot is native in ComfyUI now, so the author has stopped maintaining it and the README says so flat out. On a recent ComfyUI the built-in dual CLIP loader handles the current formats, and this node's useful life is really about loading older quantized encoder pairs that were already made with the convert_to_quant toolchain.