GGUF Convertor (Reverse)

Every other conversion node in this pack goes safetensors → gguf, shrinking a model down. This one runs the tape backward: it takes a .gguf file and turns it into .safetensors. It's a genuinely niche tool - you'll know if you need it, because you'll have a specific reason to want a gguf-only release back in a format other tools can read.

What it's actually for

Not everything ships as safetensors. If a model's only available gguf release, and you need the safetensors form for a tool that doesn't understand gguf at all, or you want to re-quantize it differently than the original conversion did, this is how you get there. It's a utility for a specific situation, not something you'll wire into a normal generation workflow.

How it works

Standalone, like the rest of this pack's conversion tools - no inputs or outputs to wire into your graph. Select the gguf file, hit Queue, watch the console. The converted safetensors file lands in ComfyUI/output.

Inputs and outputs

One input: select_gguf, a dropdown of gguf files it can find. No graph output - it's a true output node (is_output_node: true): the result is a file on disk, not a value handed to another node.

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 gguf file you want reverse-converted into ComfyUI/models/diffusion_models so it shows up in the dropdown.

The gotcha you need to know before you use the output

The README is explicit about this and it's the one thing that'll trip you up if you skip it: the reverse-converted safetensors file doesn't contain CLIP or VAE. It's not a usable standalone checkpoint - you can't load it the way you'd load a normal all-in-one safetensors model and expect it to work. Instead, drop it into diffusion_models and load it with a diffusion-model-only loader (ComfyUI's own "Load Diffusion Model," or this pack's own LoaderGGUF-style loading pattern if you're treating it like a split gguf model), then pair it with separately loaded CLIP and VAE - exactly the way you'd handle any split diffusion-only model.

Common issues

Treating the output like a full checkpoint. This is the mistake to avoid - if you try to load the reversed file expecting it to behave like a complete model with its own CLIP and VAE baked in, it won't work. Load it as a diffusion-model-only file and supply CLIP and VAE separately.

Not sure why you'd need this at all. Fair - most people never will. This exists for the specific case of needing a gguf-only release in safetensors form for a tool, script, or workflow that can't read gguf directly. If that's not your situation, you almost certainly don't need this node.

No community troubleshooting trail. Like this pack's other conversion utilities, there's essentially no public discussion of this specific node beyond the README - it's low-traffic enough that you're working from the documentation alone if something doesn't behave as expected.