Load UNET Name

The pattern repeats one more time: pick a file from a dropdown, get its filename back as a STRING, load nothing. No diffusion model gets pulled into VRAM. This one targets the "UNET" slot specifically - the diffusion model file on its own, separate from a bundled checkpoint.

Why this slot exists separately from a checkpoint

A full checkpoint bundles the diffusion model, the text encoder, and the VAE into one file. Modern large models increasingly ship unbundled instead - you load the diffusion weights, the text encoder, and the VAE as three separate files. This split matters most once you're running quantized weights: GGUF conversions (the format that made Flux-class models fit on consumer cards) and fp8 variants are typically distributed as standalone diffusion-model files rather than full checkpoints, because it's specifically the transformer-style diffusion weights that tolerate aggressive quantization well - the encoder and VAE are handled separately, often at different precision. So "UNET" here really means "the diffusion model file," in whatever precision or quantization you've got sitting in your models folder.

Once you're juggling a few of those - an fp8 version for speed, a Q8 GGUF for quality, maybe a distilled variant - knowing which exact one produced a given test render is genuinely useful, and that's the practical case for this node: pick it from the real dropdown (no chance of a typo'd filename), then route the string into a comparison-grid axis or a filename label. Per the pack's own README, this whole family is built as secondary support for XY Plot grid comparisons using qq-nodes-comfyui.

How it works

The unet_name field is a dropdown populated dynamically from your diffusion-model files - the same live scan a real UNET/diffusion-model loader does - rather than a fixed list. Select a file and the node converts your selection to text.

What to set, what it outputs

One field: unet_name, populated from whatever's in your local models folder. The single output is a STRING - tooltip "The unet name" - ready for a grid axis, a filename template, or anywhere else downstream that wants the model identified as text.

Installing it

ComfyUI Manager: search ComfyUI-ReplenishNodes, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/hben35096/ComfyUI-ReplenishNodes

Nothing to download for the node itself.

Where people get tripped up

An empty dropdown just means ComfyUI hasn't found any diffusion-model files in the expected folder yet - drop the .safetensors or .gguf file in ComfyUI/models/diffusion_models/ (or wherever your extra_model_paths.yaml maps it) and restart or refresh.

And the recurring gotcha for this whole node family applies here too: the output is text, not a loaded model, so it won't connect to anything expecting a live MODEL. If you actually want to run the diffusion model, you need the real loader - a stock Load Diffusion Model node, or a GGUF-aware loader if you're pointing at a quantized file. This node only ever produces the label.