ComfyUI-TiledVaeLite

Tiled VAE decode node for ComfyUI

Modified from https://github.com/Lightricks/ComfyUI-LTXVideo

Faster than the ComfyUI core tiled VAE decode node, and less clutter to the node tree than the full LTXV package. This adds only two nodes under the latent category. DecoderNoise has not been modified but it seemed fitting to include.

Modifications compared to the LTXV tiled VAE decode node:

• Progress bar
• 2x2 tiles and overlap 4 (32 or 64 pixels) by default when adding the node
• Actual logging instead of print so it doesn't mess up the ComfyUI bottom panel log
• Changed formatting of the log messages a little and added time measurements per tile and total

Can be useful for decoding video latents such as those generated by Wan 2.2 5B TI2V or HunyuanVideo 1.5 both of which have big and slow VAEs, or any large image, with low VRAM.