⬡ Checkpoint Loader

Load a standard SD 1.5 / SDXL checkpoint with optional external VAE.

Simple Version

Inputs

Name	Type	Required	Default	Description
ckpt_name	COMBO	✅	—	Select a checkpoint file from models/checkpoints/
vae_name	COMBO	✅	Baked VAE	External VAE, or use the one embedded in the checkpoint

Outputs

Name	Type	Description
model	MODEL	The loaded diffusion model
clip	CLIP	Text encoder for prompt processing
vae	VAE	Variational Autoencoder for image encoding/decoding
model_name	STRING	Checkpoint filename for downstream display

When to use

Use this node when you want direct access to MODEL/CLIP/VAE outputs for wiring to standard ComfyUI nodes. For a fully bundled approach, use the Advanced version below.

---

Advanced Version

All-in-one loader: checkpoint + VAE + LoRA stack + prompts + generation settings → single UME_BUNDLE output.

Inputs

Name	Type	Required	Default	Description
ckpt_name	COMBO	✅	—	Checkpoint file
vae_name	COMBO	✅	Baked VAE	External VAE override
clip_skip	INT	✅	-1	CLIP layers to skip (-1 = none, -2 = common for anime)
positive	STRING	✅	—	Positive prompt (what you want)
negative	STRING	✅	—	Negative prompt (what to avoid)
width	INT	✅	512	Output width (64–8192, step 8)
height	INT	✅	768	Output height (64–8192, step 8)
batch_size	INT	✅	1	Images per batch (1–64)
steps	INT	✅	20	Sampling steps (1–200)
cfg	FLOAT	✅	7.0	CFG scale (0–50)
sampler_name	COMBO	✅	—	Sampling algorithm
scheduler	COMBO	✅	—	Noise schedule
seed	INT	✅	0	Random seed for reproducibility
denoise	FLOAT	✅	1.0	Denoising strength (1.0 = full, <1.0 = img2img)
lora_stack	LORA_STACK	❌	—	Optional LoRA stack from a LoRA Block node

Outputs

Name	Type	Description
model_bundle	UME_BUNDLE	Complete bundle ready for KSampler

Self-contained node

The Advanced loader is a self-contained workflow-in-a-node. It loads the model, applies LoRAs, encodes prompts, and packs everything into a single bundle. Ideal for quick setups.