Get Model Name (Texturaizer)

This one's plumbing, and it's honest plumbing: it retrieves the diffusion model name from Blender's exported scene data (or a directory you point it at) so the rest of your workflow doesn't need a hardcoded model dropdown. If you're rendering from Blender and switching checkpoints per-scene inside the addon's preferences, this is what carries that choice into ComfyUI without you having to touch the graph.

How it works

Same read as every other Texturaizer "Get" node - it looks at directory_optional, or falls back to the addon's global directory, or takes an already-loaded dictionary via data_optional if you've chained it from something like Get Json Data upstream. What makes this one different is that it doesn't hand back one model name, it hands back several, because "the model" means different things depending on how your checkpoint is structured.

The outputs that matter

• checkpoint_name - for a single-file checkpoint (the classic all-in-one .safetensors with model, CLIP and VAE baked together).
• diffusion_name / unet_name - for a split setup, where the diffusion model loads on its own and CLIP/VAE come from separate files (the norm on Flux-class models).
• model_type_idx - an integer the tooltip calls "for selecting the proper model." Read: it's meant to drive a switch elsewhere in your graph that decides which of the two loading paths above actually runs, based on what Blender says you configured.
• weight_dtype - a dropdown of default, fp8_e4m3fn, fp8_e4m3fn_fast, or fp8_e5m2. This is the exact same weight-precision choice you'd see on ComfyUI's stock UNETLoader. fp8 halves the VRAM a large diffusion model needs versus fp16 with close to no visible quality loss - it's the default move on 12–24GB cards for anything Flux-sized and up, so don't be surprised Texturaizer surfaces it directly rather than assuming full precision.
• data_hash - debug-only fingerprint, per the author's tooltip; not something you wire anywhere.

You'll wire checkpoint_name or unet_name/diffusion_name into your loader of choice (Load Checkpoint, or a UNET/CLIP/VAE split loading trio), and weight_dtype into the loader's own dtype input if it has one.

Installing it

ComfyUI Manager → search Texturaizer → Install → restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentSpaceDirective/ComfyUI-Texturaizer

then restart ComfyUI. Nothing extra to pip install per the README, and no models ship with the pack itself - this node just reports a filename, it's your existing checkpoints and diffusion models doing the work.

Where people get burned

The gsc numbers on this one suggest people land here expecting a normal model-picker node, and it isn't one - it doesn't list your installed checkpoints in a dropdown, it reads a name out of Blender's export. If you're not running the Blender addon, this node has nothing to say; leaving both inputs empty just points it at a "global directory" that doesn't exist without Blender configuring it first. And if checkpoint_name and diffusion_name/unet_name both come back non-empty in a way that confuses your graph, remember they're not alternatives you choose between manually - model_type_idx is the field that's supposed to tell you (or a switch node) which one is actually meant to be used for this render.