📦 TA Load Model (with Name)

ComfyUI has a separate loader for diffusion models, checkpoints, and GGUF files, and keeping them straight in a workflow that might use all three is busywork. TA Load Model (with Name) collapses that into one dropdown: it scans all three model folders, labels each entry with a [D], [G], or [C] prefix so you can tell what you're loading, and picks the right code path automatically.

The model_file dropdown lists every diffusion model (non-GGUF), every GGUF unet, and every checkpoint it can find. The prefix isn't decoration - the node reads it at load time and dispatches: [D] loads a diffusion model from diffusion_models, [G] loads a GGUF unet, [C] loads a full checkpoint (model + clip + vae in one file). The one extra input, weight_dtype, matters only for diffusion models: "auto" lets ComfyUI decide, or you can force fp8_e4m3fn / fp8_e5m2 if you know your model likes a specific precision.

The output set is where the "with Name" part pays off. You get model, clip, and vae - the standard triple you'd expect from a loader - plus a model_name output of the pack's custom TA_MODEL_NAME type. That's a plain string of the bare model filename, and it's the glue for the whole pack: feed it into TA Filename Generator and your output files start naming themselves after the model, or feed it to the Prompt Hub / controllers to keep metadata in one place. Nothing else in stock ComfyUI hands you the model's name as a first-class output.

Two honest caveats about what it does and doesn't load. For checkpoints, clip and vae come from inside the file. For diffusion and GGUF models, the node loads the model only - clip and vae come back empty, and you still need separate CLIP and VAE loaders (or a model preset that bundles them). And [G] entries only work if you have ComfyUI-GGUF installed; the loader reaches for its UnetLoaderGGUF, and if that's missing you get a clear error telling you so. GGUF is how most people run models that won't fit at full precision - Q8 is effectively fp16 at half the size - so the convenience is real, but the dependency is on you.

Install. The pack itself installs with no Python dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/tmode-1960/TA-ComfyUI-Nodes-Pack

or via ComfyUI Manager (search "TA ComfyUI Nodes Pack"). Restart, and it's under TA Nodes/loaders. If you plan to use [G] entries, also install ComfyUI-GGUF - the pack assumes it, doesn't ship it.

Gotchas. "No models found" in the dropdown means none of the three folders have anything in them - check your models/ layout. The v2.x pack rewrite broke v1 workflows, so old graphs need rebuilding. And the [D]/[G]/[C] prefixes are what the node keys on; if you hand-edit a preset and lose the prefix, loading stops being automatic.