VAEDecodeTiled - ComfyUI Built-in Node Documentation

The VAEDecodeTiled node decodes latent representations into images using a tiled approach to handle large images efficiently. It processes the input in smaller tiles to manage memory usage while maintaining image quality. The node also supports video VAEs by processing temporal frames in chunks with overlap for smooth transitions.

Inputs

Parameter	Description	Data Type	Required	Range
`samples`	The latent representation to be decoded into images	LATENT	Yes	-
`vae`	The VAE model used for decoding the latent samples	VAE	Yes	-
`tile_size`	The size of each tile for processing (default: 512)	INT	Yes	64-4096 (step: 32)
`overlap`	The amount of overlap between adjacent tiles (default: 64)	INT	Yes	0-4096 (step: 32)
`temporal_size`	Only used for video VAEs: Amount of frames to decode at a time (default: 64)	INT	Yes	8-4096 (step: 4)
`temporal_overlap`	Only used for video VAEs: Amount of frames to overlap (default: 8)	INT	Yes	4-4096 (step: 4)

Note: The node automatically adjusts overlap values if they exceed practical limits. If tile_size is less than 4 times the overlap, the overlap is reduced to one quarter of the tile size. Similarly, if temporal_size is less than twice the temporal_overlap, the temporal overlap is halved. The node also accounts for the VAE’s internal compression ratios when calculating tile and overlap sizes for both spatial and temporal dimensions. If the input latent is a nested batch of latents, only the first item in the batch is decoded.

Outputs

Output Name	Description	Data Type
`IMAGE`	The decoded image or images generated from the latent representation. When decoding video latents, the output is a sequence of image frames.	IMAGE

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 04136ba1abd0c74e780dc405f916a08b809630ae4f41c183049535488b40fd96

​Inputs

​Outputs

Inputs

Outputs