Load Diffusion Model+

Load Diffusion Model+ is the searchable loader for standalone diffusion models - the UNET/DiT files that modern workflows load separately from their CLIP and VAE, which is how basically every Wan, Flux, and LTX video workflow is wired. If your diffusion_models folder holds a dozen quantized variants of the same Wan, this is the node that lets you type wan fp8 and see only the ones that match.

It's part of ComfyUI-FBnodes, François Beaudry's utility pack, and it's the sibling of Load Checkpoint+ with the same filtering UX: type comma-separated terms, get an AND match against name and path, see the results grouped under headers with extensions stripped. The README's example filter is literally flux, fp8, wan, low - the vocabulary of someone whose models folder is a zoo of quantizations.

How it works

Same pattern as the whole Load Model+ family: a filter box narrows a unet_name dropdown in real time, section headers in the list are safe to click (they auto-select the first real model), and the display hides .safetensors/.ckpt/.pt/.bin/.pth extensions. The path matching is the quiet power move - low finds Wan2.2-VACE-Fun-A14B-low-noise even if its filename says nothing about low, because the folder path carries the hint.

On execution it resolves your display label back to the real model path and loads it through ComfyUI's standard diffusion-model loading, returning a MODEL plus the unet_name COMBO so downstream nodes can reference what you picked. Because it's a raw diffusion-model loader, you'll need the CLIP and VAE from separate nodes - that's normal for video workflows, where a single "checkpoint" rarely exists and the pieces are loaded independently (and often as GGUF quants).

The inputs

• filter (STRING) - search terms, comma- or space-separated, AND-matched.
• unet_name (COMBO) - the filtered dropdown.

Outputs: MODEL and unet_name (COMBO).

Installing it

ComfyUI Manager - search "ComfyUI-FBnodes" - or:

cd ComfyUI/custom_nodes
git clone https://github.com/FranckyB/ComfyUI-FBnodes.git
pip install -r ComfyUI-FBnodes/requirements.txt

Restart ComfyUI. No extra downloads for this node - the models it loads are yours already.

Where it's fiddly

The filter is the whole game, so treat it like a search engine: wan plus a precision token (fp8, bf16, gguf) is the reliable combo, and the AND behavior means one bad term zeroes the list - clear it before you panic about a "missing" model. And a warning specific to this era of video models: because high-noise and low-noise VACE models are separate files, name them distinctly (or keep them in folders) or the filter can't tell them apart. This node loads one file; the pack's VACE Stitcher expects you to pick both stages yourself.