Gigachad CLIP Loader

Gigachad CLIP Loader exists because "the CLIP model" stopped being one thing years ago. Modern checkpoints split the text encoder out of the checkpoint file, and every architecture family ships a different one - SD3's three-part encoder, LTX-Video's, Wan's, a GGUF-quantized one here, a fp8 one there. This node puts all of that behind a single interface and, uniquely in this pack, lets you load two of them at once with the dual_clip toggle.

How it works

The loader scans your clip, clip_vision, text_encoders, and unet_gguf folders and merges the results into one dropdown, so any text encoder ComfyUI can see shows up here. Then it hands you three choices per slot:

• clip_type_1 - what kind of encoder this is: stable_diffusion, stable_cascade, sd3, stable_audio, mochi, ltxv, pixart, cosmos, anima, lumina2, wan, hidream, plus a few more. This is not decoration. ComfyUI needs to know the family to load the weights with the right config, and picking the wrong one gives you garbage or a crash.
• dtype_1 - the compute dtype: default, fp16, bf16, the fp8 variants (fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2), and the NV/MX formats nvfp4 and mxfp8. This is where the node earns its keep: on an 8GB card, loading your CLIP in fp8 instead of fp16 frees real memory.
• dual_clip - flip this on and a second slot (clip_name_2, clip_type_2, dtype_2) appears so you can load two encoders and get one clip output that carries both. That's the setup for models with two text encoders, like the old SD1.5 fp16/fp16_pruned pairs or SDXL's dual text-encoder configs.

The one output is clip, which feeds straight into a prompt-encoding node.

GGUF support, with a condition

If a clip_type points at a .gguf file, the node routes through ComfyUI-GGUF's CLIPLoaderGGUF - if that pack is installed. Same pattern as the Model Loader in this suite: without ComfyUI-GGUF present, it logs a warning and falls back. So GGUF-quantized CLIPs are a "works great when you've done the prerequisite" feature, not a zero-setup one.

Where you'd use it

The most common reason to reach for a separate CLIP loader: your checkpoint already gave you a clip output, but you want a different encoder - say, a GGUF-quantized text encoder to save VRAM, or the proper LTX-Video encoder for a video workflow where the checkpoint's bundled one isn't enough. The ltxv clip type is there because LTX-Video genuinely wants its own text encoder in several setups.

Install

Standard for this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/comfyui-gigachad.git

or via ComfyUI Manager, searching comfyui-gigachad, then restart. No requirements.txt ships with the pack; the GGUF path is the only optional dependency, and it comes from installing ComfyUI-GGUF separately.

Honest verdict: it's a nicer wrapper around CLIPLoader with dual-load and dtype control that stock ComfyUI makes you chain two nodes for. If you juggle different encoder families, you'll like it. If every model you run ships its own CLIP and you never think about it, this node is a solution to a problem you don't have.