GGUF Convertor (Zero)

Most nodes in this pack load a model somebody already quantized. This one does the quantizing - the pack's own README calls it the "new flagship feature," and pitches it as converting safetensors to GGUF "without any restriction; no unsupported models anymore." That's a bold claim from the author, and it comes with an equally honest caveat right next to it: any safetensors file can be converted, whether the result actually works is, in their words, "another story."

What makes it different from the other convertor

This pack has more than one GGUF-conversion node, and it's easy to conflate them. GGUF Convertor (Alpha) - a different node, GGUFSave - was the original, model-focused converter. This one, nicknamed "Zero" in the node menu, is the newer, broader version: the README specifically says it works on text/audio encoders and VAE files too, not just the main diffusion checkpoint. If you're converting anything other than a plain checkpoint, this is the one to reach for.

How it works

It runs standalone - you don't wire anything into it or out of it. Drop a safetensors file where it can see it, select it in the node's own dropdown, hit Queue, and watch your console for progress. When it finishes, the converted .gguf file lands in ComfyUI/output, ready to move into whichever loader folder it actually belongs in (diffusion_models, text_encoders, or vae, depending on what you converted).

Inputs and outputs

One input: select_safetensors, a dropdown of safetensors files it can see. No outputs in the ComfyUI-graph sense - this is an output node in the technical sense (is_output_node: true), meaning its job is to produce a file on disk, not to hand a value to another node. Nothing downstream depends on it; it's a self-contained utility you drop on the canvas and run in isolation.

Installing it

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

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

No extra dependencies on the current version. Drop the safetensors file you want to convert into ComfyUI/models/diffusion_models so it shows up in the dropdown - that's where the node looks, regardless of whether what you're converting is actually a checkpoint, an encoder, or a VAE.

When you'd actually use this

Realistically: when nobody's already published a gguf version of the model, encoder, or VAE you want to run smaller. Community gguf conversions cover the popular releases fast - often within a day for anything major - but obscure fine-tunes, niche encoders, and anything you made yourself won't have that. This node is the local fallback for those cases.

Common issues

The conversion "works" but the output doesn't. Take the author's own caveat seriously - this tool will happily convert something that then fails to load properly, or produces broken output once you try it. Test any self-converted file in a real workflow before trusting it; don't assume conversion success means usability.

No community chatter to lean on here. This is a genuinely under-discussed corner of the pack - there's essentially no public troubleshooting trail for this specific node beyond what the README itself says, which means you're mostly on your own if something goes sideways. Keep your original safetensors file; there's no guarantee the conversion is a one-shot success.

Confusing it with the Alpha convertor. If you're getting worse results than expected, double check you're not actually running the older Alpha node out of habit - Zero is the one meant to be broader and more current.