Enviral Diffusion Model Name

Modern ComfyUI workflows increasingly load the diffusion model by itself - UNETLoader (or Load Diffusion Model) instead of a monolithic checkpoint. And like every native loader, its unet_name field is a dropdown you can't wire into. That's fine for a fixed workflow and a real obstacle for the branching "one graph, many models" setups people actually build.

Enviral Diffusion Model Name is the adapter: a dropdown of your registered diffusion models that also accepts a linked string, then outputs the exact native combo type UNETLoader expects on unet_name.

How it works

The options are enumerated from folder_paths.get_filename_list("diffusion_models"), so whatever ComfyUI can see - including entries from extra_model_paths.yaml - shows up here. Linked strings are resolved leniently: exact match, then case/path-normalized, then unique path suffix. Krea 2's checkpoint at diffusion_models/krea/Krea-2-Turbo.safetensors? A bare Krea-2-Turbo.safetensors string still resolves. If nothing matches, validation fails early with closest-match suggestions, before the loader runs.

That resolve logic is shared across the whole family of Enviral selectors, which is worth knowing: learn one, and you've learned Enviral Checkpoint Name, Enviral CLIP Type, and friends.

Input and output

• unet_name - dropdown or linked string.
• Output unet_name - native combo, wired into UNETLoader / Load Diffusion Model.

Two sockets, no other knobs.

Install

Part of the Enviral Design Node Pack. ComfyUI Manager: search "Enviral Design Node Pack". Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/EnviralDesign/comfyUI-enviral-design-node-pack

then restart. No model downloads, no Python deps.

Where it shines

Given how many current models ship as standalone diffusion files - Krea 2 Raw/Turbo, Z-Image, Flux-style splits - a "pick the model family" string that drives this node, Enviral CLIP Name, and Enviral CLIP Type together is the clean way to build one adaptive graph instead of three near-identical ones. If your loaders are static, skip it; the moment you automate model selection, this is the missing joint.