Blender Input Load Checkpoint

In a normal ComfyUI workflow, switching the base model means going back to the graph and poking a Load Checkpoint dropdown. BlenderInputLoadCheckpoint is that same loader, re-skinned so the checkpoint dropdown lives on the Blender add-on's panel instead - you change models from the viewport, mid-shot, without ever touching ComfyUI. For someone doing character or style swaps across takes in Blender, that's the difference between an annoying hop and a smooth one.

Don't overthink the mechanism: the node is literally a subclass of ComfyUI's built-in CheckpointLoaderSimple. Its INPUT_TYPES calls the parent's, then tacks on the pack's standard order and default widgets plus the optional group socket. Its execute normalizes the path and calls the parent's load_checkpoint - same loading code, same MODEL/CLIP/VAE outputs you already know. The only additions are the ones that configure how the add-on displays the loader.

Inputs and outputs

• ckpt_name - the checkpoint to load, from your ComfyUI checkpoints folder. In Blender this becomes a dropdown of every checkpoint the server sees.
• default - which checkpoint the dropdown starts on. Leave it empty to start unselected, or set a name to pin the workflow to a known model.
• order - panel position, lower first.

Outputs are the standard triple: MODEL (the denoising network), CLIP (for text-encoding your prompts), and VAE (image ↔ latent conversion). Wire them exactly like you would a normal checkpoint loader. If the workflow only needs the diffusion model itself - Flux-style setups load the UNet separately - you'll use BlenderInputLoadDiffusionModel instead; this node is for the all-in-one checkpoint case.

Installing it

Standard for the pack. ComfyUI side, ComfyUI Manager → search "ComfyUI-Blender", or:

cd ComfyUI/custom_nodes
git clone https://github.com/alexisrolland/ComfyUI-Blender

Restart ComfyUI. No extra Python dependencies, but you need a current ComfyUI - schema v3 nodes won't load on older builds, and the README is blunt about that. The Blender add-on is the other half, a ZIP from the latest release installed via Edit > Preferences > Add-ons > Install from Disk.

Where people get burned

The dropdown lists whatever's on the server's checkpoints folder - if you run ComfyUI on a remote box and Blender locally, you pick from the server's models, not your Blender machine's. That trips up a lot of people. Second, the panel only builds after an API-format export and import, and the node title is the label - "Base Model" beats "ckpt". Third, remember this is still a full checkpoint load: swapping models between takes reloads VRAM, and when Blender is sharing that GPU with ComfyUI the viewport pressure is real. The loader is free; the model you just switched to isn't.