Skip to main content
Image nodes work with final pixel-space images for loading, processing, and saving outputs.

​
Loading Images

​
LoadImage

Loads an image from the input directory. Category: image
​
image
string
required
The image file to load (supports upload)
Returns:
  • IMAGE: The loaded image in RGB format
  • MASK: Alpha channel mask (if present)
Description: Loads images from the input folder. Supports most image formats including PNG, JPEG, WebP, and animated formats. The alpha channel is extracted as a separate mask output. Supported Formats:
  • PNG (with alpha channel support)
  • JPEG
  • WebP
  • GIF (first frame only)
  • TIFF
  • BMP

​
LoadImageMask

Loads an image and extracts a specific channel as a mask. Category: mask
​
image
string
required
The image file to load
​
channel
string
required
Which channel to extract. Options: alpha, red, green, blue
Returns:
  • MASK: The extracted channel as a mask
Description: Extracts a single channel from an image to use as a mask. The alpha channel is inverted (transparent becomes white in the mask).

​
Saving Images

​
SaveImage

Saves images to the output directory. Category: image
​
images
IMAGE
required
The images to save
​
filename_prefix
STRING
default:"ComfyUI"
Prefix for the saved filename. Supports formatting like %date:yyyy-MM-dd% or %node_name.parameter%
Description: Saves images as PNG files with embedded metadata including the workflow and prompts (unless metadata is disabled). Images are saved to the output folder with sequential numbering. Filename Formatting:
  • %date:yyyy-MM-dd%: Current date
  • %date:HH-mm-ss%: Current time
  • %Empty Latent Image.width%: Reference node parameters
  • %batch_num%: Batch number for multiple images

​
PreviewImage

Displays images in the UI without saving to output. Category: image
​
images
IMAGE
required
The images to preview
Description: Shows images in the ComfyUI interface. Images are saved temporarily and deleted on server restart. Useful for debugging workflows.

​
Image Scaling

​
ImageScale

Resizes an image to specific dimensions. Category: image/upscaling
​
image
IMAGE
required
The image to scale
​
upscale_method
string
required
Interpolation method. Options: nearest-exact, bilinear, area, bicubic, lanczos
​
width
int
default:"512"
Target width. Range: 0 to 16384. Set to 0 to maintain aspect ratio.
​
height
int
default:"512"
Target height. Range: 0 to 16384. Set to 0 to maintain aspect ratio.
​
crop
string
required
Cropping method. Options: disabled, center
Returns:
  • IMAGE: Resized image
Upscale Methods:
  • nearest-exact: Fastest, pixelated look
  • bilinear: Fast, smooth but slightly blurry
  • area: Best for downscaling
  • bicubic: Good quality, balanced
  • lanczos: Highest quality, slower, sharp edges

​
ImageScaleBy

Resizes an image by a scale factor. Category: image/upscaling
​
image
IMAGE
required
The image to scale
​
upscale_method
string
required
Interpolation method
​
scale_by
float
default:"1.0"
Scale multiplier. Range: 0.01 to 8.0
Returns:
  • IMAGE: Scaled image

​
ImageUpscaleWithModel

Upscales an image using an AI upscale model. Category: image/upscaling
​
upscale_model
UPSCALE_MODEL
required
The upscale model to use (e.g., RealESRGAN, ESRGAN)
​
image
IMAGE
required
The image to upscale
Returns:
  • IMAGE: Upscaled image
Description: Uses neural network upscaling models for high-quality super-resolution. Much better quality than simple interpolation, especially for 2x or 4x upscaling. Common Models:
  • RealESRGAN_x4plus: General purpose 4x upscaling
  • RealESRGAN_x2plus: 2x upscaling
  • ESRGAN_4x: Anime/illustration focused
  • Various specialized models for faces, anime, etc.

​
Image Processing

​
ImageInvert

Inverts the colors of an image. Category: image
​
image
IMAGE
required
The image to invert
Returns:
  • IMAGE: Inverted image
Description: Inverts all colors (255 - value for each channel). Black becomes white, colors become their complementary colors.

​
ImagePadForOutpaint

Expands an image’s canvas for outpainting. Category: image
​
image
IMAGE
required
The image to expand
​
left
int
default:"0"
Pixels to add on left. Range: 0 to 16384, step 8
​
top
int
default:"0"
Pixels to add on top. Range: 0 to 16384, step 8
​
right
int
default:"0"
Pixels to add on right. Range: 0 to 16384, step 8
​
bottom
int
default:"0"
Pixels to add on bottom. Range: 0 to 16384, step 8
​
feathering
int
default:"40"
Feathering amount at edges for smooth blending. Range: 0 to 16384
Returns:
  • IMAGE: Padded image with gray fill
  • MASK: Mask marking the new areas (white = generate, black = preserve)
Description: Extends the canvas around an image for outpainting workflows. The original image is preserved in the center, and new areas are filled with neutral gray. The mask indicates which areas should be generated.

​
EmptyImage

Creates a blank colored image. Category: image
​
width
int
default:"512"
Image width. Range: 1 to 16384
​
height
int
default:"512"
Image height. Range: 1 to 16384
​
batch_size
int
default:"1"
Number of images. Range: 1 to 4096
​
color
int
default:"0"
RGB color value (hex). Range: 0 to 0xFFFFFF
Returns:
  • IMAGE: Blank colored image
Description: Creates a solid color image. Useful for backgrounds or testing.

​
Batch Operations

​
ImageBatch

Combines two images into a batch. Category: image (Deprecated)
​
image1
IMAGE
required
First image
​
image2
IMAGE
required
Second image
Returns:
  • IMAGE: Combined batch
Description: Merges two images into a single batch. If dimensions don’t match, image2 is resized to match image1. This node is deprecated; prefer using multiple inputs on nodes that support batching.

​
Workflow Examples

​
Basic Image Generation

CheckpointLoaderSimple → MODEL, CLIP, VAE
CLIPTextEncode (positive) → CONDITIONING
CLIPTextEncode (negative) → CONDITIONING  
EmptyLatentImage → LATENT
KSampler → LATENT
VAEDecode → IMAGE
SaveImage

​
Image to Image

LoadImage → IMAGE, MASK
VAEEncode → LATENT
KSampler (denoise < 1.0) → LATENT
VAEDecode → IMAGE
SaveImage

​
Upscaling with Model

[Generate base image]
VAEDecode → IMAGE
UpscaleModelLoader → UPSCALE_MODEL
ImageUpscaleWithModel → IMAGE (high res)
VAEEncode → LATENT
KSampler (denoise 0.3-0.5) → LATENT (refined)
VAEDecode → IMAGE
SaveImage

​
Outpainting

LoadImage → IMAGE
ImagePadForOutpaint → IMAGE, MASK
VAEEncode → LATENT
SetLatentNoiseMask → LATENT (with mask)
KSampler → LATENT
VAEDecode → IMAGE
SaveImage

​
Image Formats

ComfyUI internally uses images in the format:
  • Shape: [batch, height, width, channels]
  • Data type: float32
  • Range: 0.0 to 1.0
  • Color space: RGB (not BGR)
Images are automatically converted to/from this format when loading and saving.

​
Tips

​
Memory Usage

  • Pixel-space operations use more memory than latent operations
  • For large images, consider working in latent space when possible
  • Use tiled VAE encoding/decoding for very large images

​
Quality Settings

  • PNG compression level is set to 4 by default (balance of size/speed)
  • Metadata is embedded in saved PNGs (disable with --disable-metadata flag)
  • Preview images use lower compression for faster display

​
File Organization

  • Input images go in input/ directory
  • Output images save to output/ directory
  • Preview images save to temp/ directory
  • Use filename_prefix to organize outputs into subdirectories

Build docs developers (and LLMs) love

Get started for freeTalk to us