HyperTile - ComfyUI Built-in Node Documentation

HyperTile applies a tiling technique to the attention mechanism inside diffusion models to reduce memory usage during image generation. It splits the latent space into smaller tiles, processes the attention for each tile separately, and then reassembles the results. This makes it possible to work with larger image sizes without running out of memory.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The diffusion model to apply the HyperTile optimization to	MODEL	Yes	-
`tile_size`	The target tile size for processing (default: 256). Internally, the value is clamped to a minimum of 32 and then divided by 8 to obtain the effective tile size.	INT	Yes	1 - 2048
`swap_size`	Controls how the tiles are rearranged during processing to improve efficiency. Larger values allow more variation in tile sizes (default: 2)	INT	Yes	1 - 128
`max_depth`	The maximum depth level (resolution scale) to apply tiling. A value of 0 applies tiling only at the highest resolution (default: 0)	INT	Yes	0 - 10
`scale_depth`	When enabled, the tile size is scaled proportionally at deeper depth levels. This can help maintain quality at lower resolutions (default: False)	BOOLEAN	Yes	True / False

Note: tile_size, swap_size, max_depth, and scale_depth are marked as advanced inputs, so they are only shown when advanced options are enabled in the interface.

Outputs

Output Name	Description	Data Type
`model`	The modified model with HyperTile optimization applied	MODEL

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): fb2fa29a403b6b7de7d5263240cc51a74126078457a3ff9ea63aeded45b9b74a

​Inputs

​Outputs

Inputs

Outputs