KJNodes/vaegenerated

VAELoader KJ(VAELoaderKJ)

This node not only supports loading standard VAE model files but also specifically integrates support for various TAESD (Tiny AutoEncoder for Stable Diffusion) approximation models, including dedicated versions for images (e.g., taesd, taesdxl) and videos (e.g., taehv, lighttaew2_2).

VAE Loader KJ

VAE
vae_name
COMBO
device
COMBO
weight_dtype
COMBO
KJNodes

The VAELoader KJ node not only supports loading standard VAE model files but also specifically integrates support for various TAESD (Tiny AutoEncoder for Stable Diffusion) approximation models, including dedicated versions for images (e.g., taesd, taesdxl) and videos (e.g., taehv, lighttaew2_2).

VAELoader KJ is a node for loading Variational Autoencoder (VAE) models. It can read and prepare VAE files from a specified model folder for subsequent image generation or processing workflows. This node is a key component connecting model files with the generation workflow, ensuring the correct VAE is loaded into the computational environment.

Node Functionality

This node not only supports loading standard VAE model files but also specifically integrates support for various TAESD (Tiny AutoEncoder for Stable Diffusion) approximation models, including dedicated versions for images (e.g., taesd, taesdxl) and videos (e.g., taehv, lighttaew2_2). Its core mechanism involves scanning the vae and vae_approx directories to automatically combine matching encoder and decoder files to construct a complete VAE instance. It also allows users to specify the numerical precision and device location of the model in memory.

Node Parameter Description - VAELoader KJ

Connection Inputs (Inputs)

Parameter NameData TypeRequiredDefault ValueValue Range/OptionsDescription
vae_nameCOMBOYes--vae_name (input parameter): Selects the name of the VAE model to load. The name list is automatically generated by the node by scanning the vae and vae_approx folders.

Control Parameters (Parameters)

Parameter NameData TypeRequiredDefault ValueValue Range/OptionsDescription
deviceCOMBOYesmain_device"main_device", "cpu"device (input parameter): Specifies the computing device for loading the VAE model. Options are "main_device" (typically GPU) or "cpu".
weight_dtypeCOMBOYesbf16"bf16", "fp16", "fp32"bf16, fp16, or fp32, used to balance computational precision and memory usage.

Output

Parameter NameData TypeDescription
VAEVAEVAE (output parameter): Outputs the loaded and configured VAE model object for subsequent image decoding or encoding steps.

Usage Scenarios

In a typical text-to-image or image-to-image workflow, you would use this node after loading the main model (e.g., Checkpoint). For example, to generate an image with richer details, you can connect this node's VAE output to the corresponding input of a KSampler node, thereby applying a specific VAE to decode latent space features during the sampling process.

Notes

Please note that whether the node can successfully load combined models like taesd or taesdxl depends on the presence of corresponding encoder and decoder files in the vae_approx folder. Additionally, choosing lower-precision data types like bf16 can save VRAM but may, in rare cases, affect numerical stability.

The VAELoader KJ node is from the ComfyUI-KJNodes node package.

Comments

Sign in with GitHub to join the discussion.

Loading comments…