Model
Trends
.ai
Model
Trends
.ai
the best open source ai models

Load Checkpoint node in ComfyUI

Load Checkpoint (CheckpointLoaderSimple) opens a .safetensors or .ckpt model file and hands out the three parts every workflow needs: the diffusion MODEL, the CLIP text encoder and the VAE.
Node pack: ComfyUI core
Category: Loaders

What Load Checkpoint does

Load Checkpoint is the first node in almost every ComfyUI graph. It reads one checkpoint file from the models/checkpoints folder and splits it into the three objects that the rest of the graph consumes: MODEL (the UNet or transformer that denoises), CLIP (the text encoder that turns prompts into conditioning) and VAE (the encoder/decoder between pixels and latent space).
The node has a single dropdown, ckpt_name, listing every file ComfyUI found in the checkpoints folder (and any extra paths from extra_model_paths.yaml). Pick a file, connect the three outputs, and the model family is set for the whole graph. SD 1.5, SDXL, Pony, Illustrious and most merged checkpoints load through this node. Flux, SD3 and video models usually ship as separate UNet, text encoder and VAE files and use Load Diffusion Model, DualCLIPLoader and Load VAE instead.

Inputs

Name
Type
What it is
ckpt_name
COMBO
The checkpoint file to load, chosen from models/checkpoints. The list is read when the page loads; new files appear after a refresh of the node list.

Outputs

Name
Type
What it is
MODEL
MODEL
The denoising network. Connect it to KSampler, LoRA loaders, ControlNet or model patch nodes.
CLIP
CLIP
The text encoder. Connect it to CLIP Text Encode for the positive and negative prompts.
VAE
VAE
The pixel/latent converter. Connect it to VAE Decode and VAE Encode.
AD
Skip the setup: ComfyUI workflows preinstalled on the cloud
BitVector runs ready-made ComfyUI workflows on its own GPUs. No install, no missing nodes, no red boxes. Open it on your phone or laptop and generate in a minute.

How to use Load Checkpoint

  1. Copy the checkpoint into ComfyUI/models/checkpoints (or a folder listed in extra_model_paths.yaml).
  2. Add the node with double-click search: type "Load Checkpoint".
  3. Choose the file in ckpt_name.
  4. Wire MODEL to the KSampler model input, CLIP to both CLIP Text Encode nodes, VAE to VAE Decode.
  5. Match the rest of the graph to the model family: 512 or 768 latents for SD 1.5, 1024 for SDXL.

Settings and tips

  • Prefer .safetensors files; .ckpt files are pickled Python and can run code when loaded.
  • If a checkpoint was published without a baked-in VAE, colours look washed out. Add Load VAE with the matching VAE and use that instead of the checkpoint VAE output.
  • One Load Checkpoint can feed several samplers. You do not need a second loader for a hires pass.
  • Loading is cached: switching back and forth between two checkpoints is fast after the first load, but each one stays in RAM.
  • Use Load LoRA between this node and the sampler, never before it; the LoRA patches the MODEL and CLIP that come out of here.

Troubleshooting Load Checkpoint

ckpt_name shows "undefined" or the dropdown is empty

Why it happens
ComfyUI did not find any checkpoint in models/checkpoints when it started, or the workflow was saved with a file name that does not exist on this machine.

How to fix it
Put at least one .safetensors file in ComfyUI/models/checkpoints, press R to refresh the node definitions, then pick the file again. For shared workflows, choose any local checkpoint of the same family.

Error while deserializing header: HeaderTooLarge or MetadataIncompleteBuffer

Why it happens
The file is not a complete safetensors file. Usually the download stopped early, or an HTML error page was saved with a .safetensors name.

How to fix it
Check the file size against the download page and download it again. A healthy SDXL checkpoint is around 6.5 GB; a file of a few kilobytes is an error page.

Images come out as noise or grey squares with a valid checkpoint

Why it happens
The checkpoint family does not match the rest of the graph: an SDXL model with a 512x512 latent and SD 1.5 settings, or a Flux UNet loaded through Load Checkpoint.

How to fix it
Use the resolution and sampler settings of the model family (SDXL: 1024x1024, cfg 5-8). For Flux, SD3 and Wan use Load Diffusion Model plus the separate CLIP and VAE loaders.

Clip missing: ["text_projection.weight"] or "clip missing" warnings in the console

Why it happens
The checkpoint was saved without part of its text encoder. ComfyUI fills the gap with defaults, which can shift the style slightly.

How to fix it
It is a warning, not a failure. If results differ from the model page, load the CLIP from a complete checkpoint of the same family and use that CLIP output instead.

Out of memory as soon as the checkpoint loads

Why it happens
The whole checkpoint is moved to the GPU and there is less free VRAM than the model needs, often because another program or a second ComfyUI tab holds memory.

How to fix it
Close other GPU programs, start ComfyUI with --lowvram or --normalvram, or load the UNet in fp8 through Load Diffusion Model. 6 GB cards can run SDXL with --lowvram.

AD
Run this workflow from your phone
Every workflow on this page is preinstalled on BitVector with the models and custom nodes already in place. Pick one, type a prompt, done. Zero setup, nothing to download.

Questions about Load Checkpoint

Is Load Checkpoint the same as CheckpointLoaderSimple?

Yes. CheckpointLoaderSimple is the internal class name; the menu shows it as Load Checkpoint. The older CheckpointLoader with a config file dropdown is deprecated.

Can I load a LoRA with this node?

No. LoRAs go into Load LoRA, which takes the MODEL and CLIP from this node and returns patched versions.

Why do Flux models not appear in the list?

Flux is distributed as a UNet file that belongs in models/unet (or models/diffusion_models). Load it with Load Diffusion Model and add DualCLIPLoader and Load VAE.

Related nodes

Load Diffusion Model
ComfyUI core
Load Diffusion Model (UNETLoader) loads a bare denoising network such as Flux, SD3.5, Wan or HunyuanVideo from models/diffusion_models, with a weight_dtype option for fp8 to save VRAM.
Load LoRA
ComfyUI core
Load LoRA (LoraLoader) applies a LoRA file to the MODEL and CLIP with separate strengths, so a style, character or concept can be added to any checkpoint without merging.
Load VAE
ComfyUI core
Load VAE (VAELoader) loads a standalone VAE file from models/vae so a checkpoint with a missing or weak VAE decodes clean colours, and so split models like Flux and Wan get their decoder.
CLIP Text Encode (Prompt)
ComfyUI core
CLIP Text Encode turns a text prompt into CONDITIONING using the model text encoder. One node holds the positive prompt, a second one the negative prompt.
KSampler
ComfyUI core
KSampler runs the denoising loop: it takes the model, prompts and a latent and produces the finished latent image, controlled by seed, steps, cfg, sampler, scheduler and denoise.
AD
Tired of fixing nodes? Let the cloud do it
BitVector keeps hundreds of ComfyUI workflows installed, updated and tested on fast cloud GPUs. No Python, no CUDA errors, no VRAM limits. Works from any browser.

More ComfyUI nodes

ComfyUI Manager
ComfyUI-Manager
ComfyUI Manager is the extension that installs, updates and fixes custom node packs and models from inside the interface, resolves missing nodes in imported workflows and snapshots your setup.
Save Image
ComfyUI core
Save Image writes the IMAGE tensor to ComfyUI/output as a PNG, with the whole workflow embedded in the file metadata so the picture can be dragged back into ComfyUI to restore the graph.
VAE Decode
ComfyUI core
VAE Decode converts the sampled LATENT into a pixel IMAGE with the VAE; VAE Decode (Tiled) does the same in tiles for very large images.
Empty Latent Image
ComfyUI core
Empty Latent Image creates the blank latent canvas (width, height, batch_size) that text-to-image sampling starts from; dimensions must be multiples of 8 and match the model family.
Load Image
ComfyUI core
Load Image reads a picture from the ComfyUI input folder (or an upload) and outputs an IMAGE tensor plus a MASK taken from the alpha channel, the starting point for img2img, inpainting, ControlNet and IP-Adapter graphs.
Apply ControlNet
ComfyUI core
Apply ControlNet attaches a guide image and a loaded ControlNet to the positive and negative CONDITIONING with strength and start/end percentages, so the sampler follows edges, depth or pose from the guide.