VAEEncodeBatch

Documentation

• Class name: VAEEncodeBatch
• Category: Bmad
• Output node: False

The VAEEncodeBatch node is designed to encode a batch of images into their latent representations using a Variational Autoencoder (VAE). It processes multiple images sequentially, leveraging a VAE model to transform each image into a latent space vector, and then concatenates these vectors to form a batch of latent representations.

Input types

Required

• inputs_len
  • Specifies the number of images to encode into their latent representations. It determines the size of the batch that will be processed by the node.
  • Comfy dtype: INT
  • Python dtype: int
• vae
  • The Variational Autoencoder (VAE) model used for encoding images into their latent representations. It is a crucial component that defines the encoding mechanism.
  • Comfy dtype: VAE
  • Python dtype: object

Output types

• latent
  • Comfy dtype: LATENT
  • The latent representation of the encoded images. It is a batch of vectors in the latent space, representing the input images.
  • Python dtype: Dict[str, torch.Tensor]

Usage tips

• Infra type: GPU
• Common nodes: unknown

Source code

class VAEEncodeBatch:

    @classmethod
    def INPUT_TYPES(cls):
        return {"required": {
            "inputs_len": ("INT", {"default": 3, "min": 2, "max": 32, "step": 1}),
            "vae": ("VAE",)
        }}

    RETURN_TYPES = ("LATENT",)
    FUNCTION = "encode"
    CATEGORY = base_category_path

    def encode(self, inputs_len, vae, **kwargs):
        vae_encoder = nodes.VAEEncode()

        def get_latent(input_name):
            pixels = kwargs[input_name]
            pixels = vae_encoder.vae_encode_crop_pixels(pixels)
            return vae.encode(pixels[:, :, :, :3])

        latent = get_latent("image_1")
        for r in range(1, inputs_len):
            latent = torch.cat([latent, get_latent(f"image_{r + 1}")], dim=0)

        return ({"samples": latent},)