GGUF AudioEncoder Loader

Most of the GGUF conversation is about images and video - that's genuinely where the ecosystem's attention lives, and the community-facing GGUF tooling most people know (city96's pack included) is scoped to those two modalities. This node is one of the places calcuis's gguf pack goes further: it loads a quantized or plain safetensors audio-encoder file for ComfyUI's audio-generation pipelines, the same way its sibling nodes handle image and video encoders.

Why it exists

Audio generation in ComfyUI (think Stable Audio-family workflows - you'll notice stable_audio sitting right alongside the image and video architectures in this pack's other CLIP-type dropdowns) needs its own encoder, same conceptual role as a text encoder for image models: it converts input into the representation the audio diffusion model actually conditions on. As with image/video, that encoder file can be large enough that quantizing it matters on a constrained card - and this loader is what reads a gguf version of one.

How it works

Same pattern as this pack's GGUF VAE Loader: it accepts either a .gguf or a .safetensors audio-encoder file through one loader, so you're not stuck matching a specific node to a specific file format. Point it at a file, get back an encoder object your audio-generation graph can use.

Inputs and outputs

One input: audio_encoder_name, a dropdown of whatever audio-encoder files it finds. One output: AUDIO_ENCODER, which wires into whichever audio-generation node in your graph expects that type - this loader's job stops at handing off the encoder object, same as any of the other loaders in this pack.

Installing it

ComfyUI Manager: search gguf, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/calcuis/gguf

No extra pip installs needed on the current version. The pack's README doesn't spell out a dedicated folder for audio-encoder files the way it does for diffusion models, text encoders, VAE, ControlNet, and LoRA - the safe default is ComfyUI's standard models/audio_encoders location, the same place a native audio encoder loader would expect to find them. If your dropdown comes up empty, that folder is the first thing to check.

Common issues

This node feels obscure because it is, right now. Zero recorded search traffic on it isn't a fluke - audio generation is a smaller corner of the ComfyUI ecosystem than image or video, and GGUF tooling for it specifically is even smaller than that. If you landed here, you're almost certainly already deep into a specific audio workflow rather than browsing.

Getting a gguf version of your audio encoder in the first place. Unlike image models, where community-quantized gguf files are everywhere on HuggingFace, audio encoders in gguf format are much rarer to find pre-made. This pack's own conversion tools (GGUF Convertor Zero, specifically - the one that explicitly claims to handle encoders, not just checkpoints) are the practical way to produce one yourself from a safetensors audio encoder, if nobody's published a ready-made quant of the one you need.

Empty dropdown. File's not in the folder ComfyUI's scanning, or it was added after the app was already running - restart to pick up new files, and double check the folder path if you're not sure ComfyUI's looking where you put it.

Mismatch with the rest of your audio graph. As with any encoder, this node only guarantees it can load a file - pairing it with the wrong audio diffusion model will still run, it just won't sound right. Match the encoder to whatever your specific audio model's documentation calls for.