GGUF Convertor (Alpha)

This is the older of this pack's two safetensors-to-GGUF converters - labeled "Alpha" right there in the node menu, which is the author's own way of telling you it came first and hasn't been superseded so much as sat alongside its newer sibling, GGUF Convertor (Zero). If you just want to turn a diffusion checkpoint into a smaller gguf file, this node still does exactly that.

What it does

Converts a safetensors diffusion model to .gguf, full stop. Unlike Zero, which the README explicitly says also handles text/audio encoders and VAE files, Alpha is scoped to the main checkpoint. If that's genuinely all you're converting, it's a perfectly fine tool for the job - just know Zero exists and covers more ground if you later need to convert something that isn't a checkpoint.

How it works

It runs completely standalone, same as the rest of this pack's conversion tools - nothing wired in, nothing wired out. Select your file, hit Queue, and the console shows progress while it works. The finished .gguf file lands in ComfyUI/output.

Inputs and outputs

One input: select_safetensors, a dropdown pulled from your safetensors files. No graph-level output - this is a true output node (is_output_node: true): its result is a file written to disk, not a value passed to the next node. There's nothing downstream to wire it into.

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. Put the safetensors file you want to convert in ComfyUI/models/diffusion_models so this node can see it.

The one real technique the README describes for this node

There's a specific trick the author calls out for getting a smaller, theoretically-faster-loading result: cut your source model in half first with TENSOR Cutter (bf16 down to fp8), then run the trimmed fp8 file through this converter instead of the original bf16. You end up with a gguf file that's roughly the same size as converting straight from bf16, but with fewer tensors packed inside, which the author says should load faster - with the honest caveat that there's no guarantee, and you should expect a real, visible quality hit from the extra cutting step before you even get to the gguf conversion. Test before you commit to a workflow built around it.

Common issues

Not sure whether to use this or Zero. Reach for this one only if you're specifically converting a plain checkpoint and have no interest in the newer node's broader scope. For anything involving encoders or VAE, or if you just want the more current tool, use Zero instead.

Conversion succeeds, output doesn't work well. As with its sibling, there's no guarantee a converted file behaves identically to a proper community-quantized release - test it in a real generation before trusting it for anything you care about.

No community reports to draw on. This is a low-traffic, self-contained utility node with essentially no public discussion trail beyond the README itself - there isn't a body of known failure modes to hand you beyond what's documented here.