Skip to main content
Loader nodes load models and resources that are used throughout your workflow. These nodes typically run once at the beginning of a workflow.

​
Checkpoint Loaders

​
CheckpointLoaderSimple

Loads a diffusion model checkpoint. Checkpoints contain the MODEL, CLIP, and VAE needed for image generation. Category: loaders
​
ckpt_name
string
required
The name of the checkpoint (model) to load from the checkpoints directory
Returns:
  • MODEL: The diffusion model used for denoising latents
  • CLIP: The CLIP model used for encoding text prompts
  • VAE: The VAE model for encoding/decoding images to/from latent space
Description: Loads a diffusion model checkpoint. Diffusion models are used to denoise latents. The checkpoint is automatically detected and configured.

​
unCLIPCheckpointLoader

Loads an unCLIP model checkpoint, which includes CLIP Vision support. Category: loaders
​
ckpt_name
string
required
The checkpoint file to load
Returns:
  • MODEL: The diffusion model
  • CLIP: The CLIP text encoder
  • VAE: The VAE model
  • CLIP_VISION: The CLIP vision model for image conditioning

​
LoRA Loaders

​
LoraLoader

Applies a LoRA (Low-Rank Adaptation) to modify the model and CLIP behavior. Category: loaders
​
model
MODEL
required
The diffusion model the LoRA will be applied to
​
clip
CLIP
required
The CLIP model the LoRA will be applied to
​
lora_name
string
required
The name of the LoRA file to load
​
strength_model
float
default:"1.0"
How strongly to modify the diffusion model. Range: -100.0 to 100.0. Can be negative.
​
strength_clip
float
default:"1.0"
How strongly to modify the CLIP model. Range: -100.0 to 100.0. Can be negative.
Returns:
  • MODEL: The modified diffusion model
  • CLIP: The modified CLIP model
Description: LoRAs are used to modify diffusion and CLIP models, altering the way latents are denoised. This is commonly used for applying styles. Multiple LoRA nodes can be chained together.

​
LoraLoaderModelOnly

Applies a LoRA only to the model, not CLIP. Category: loaders
​
model
MODEL
required
The diffusion model to apply the LoRA to
​
lora_name
string
required
The name of the LoRA file
​
strength_model
float
default:"1.0"
LoRA strength. Range: -100.0 to 100.0
Returns:
  • MODEL: The modified diffusion model

​
VAE Loaders

​
VAELoader

Loads a VAE model separately from a checkpoint. Category: loaders
​
vae_name
string
required
The VAE file to load. Options include standard VAEs and approximate VAEs (TAESD variants)
Returns:
  • VAE: The loaded VAE model
Available VAE Types:
  • Standard VAE models (.safetensors, .pt files)
  • taesd: Tiny AutoEncoder for SD 1.x (fast preview)
  • taesdxl: Tiny AutoEncoder for SDXL (fast preview)
  • taesd3: Tiny AutoEncoder for SD3 (fast preview)
  • taef1: Tiny AutoEncoder for Flux.1 (fast preview)
  • Video VAEs: taehv, lighttaew2_2, lighttaew2_1, etc.
  • pixel_space: Bypass VAE encoding/decoding entirely

​
CLIP Loaders

​
CLIPLoader

Loads a CLIP text encoder model separately. Category: advanced/loaders
​
clip_name
string
required
The CLIP model file to load from the text_encoders directory
​
type
string
required
The CLIP architecture type. Options: stable_diffusion, stable_cascade, sd3, stable_audio, mochi, ltxv, pixart, cosmos, etc.
​
device
string
default:"default"
Device to load on. Options: default, cpu
Returns:
  • CLIP: The loaded CLIP model

​
DualCLIPLoader

Loads two CLIP models for architectures that use dual text encoders. Category: advanced/loaders
​
clip_name1
string
required
First CLIP model file
​
clip_name2
string
required
Second CLIP model file
​
type
string
required
Dual CLIP type. Options: sdxl, sd3, flux, hunyuan_video, hidream, etc.
​
device
string
default:"default"
Device to load on
Returns:
  • CLIP: The combined dual CLIP model
Common Recipes:
  • SDXL: clip-l + clip-g
  • SD3: clip-l + clip-g or clip-l + t5 or clip-g + t5
  • Flux: clip-l + t5

​
Advanced Loaders

​
UNETLoader

Loads a diffusion model (UNET) separately. Category: advanced/loaders
​
unet_name
string
required
The UNET model file to load
​
weight_dtype
string
default:"default"
Weight precision. Options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2
Returns:
  • MODEL: The loaded diffusion model

​
ControlNetLoader

Loads a ControlNet model for guided image generation. Category: loaders
​
control_net_name
string
required
The ControlNet model file to load
Returns:
  • CONTROL_NET: The loaded ControlNet model

​
CLIPVisionLoader

Loads a CLIP Vision model for image conditioning. Category: loaders
​
clip_name
string
required
The CLIP Vision model file to load
Returns:
  • CLIP_VISION: The loaded CLIP Vision model

​
Model-Specific Loaders

​
StyleModelLoader

Loads a style model for style transfer. Category: loaders
​
style_model_name
string
required
The style model file to load
Returns:
  • STYLE_MODEL: The loaded style model

​
UpscaleModelLoader

Loads an upscale model for image super-resolution. Category: loaders
​
model_name
string
required
The upscale model file to load from the upscale_models directory
Returns:
  • UPSCALE_MODEL: The loaded upscale model

Build docs developers (and LLMs) love

Get started for freeTalk to us