A VAE for Wan2.1 and Krea2...
Inspired by JoeyGambino and his Z-Image + H3/LTX merge/grafting process. Using one of the highest benchmarked VAE currently available as the mathematical basis for modifications to the existing WAN VAE and it's 2x Upscale variant.
For the Krea2 Upscale version use this node pack to run it:
https://github.com/spacepxl/ComfyUI-VAE-Utils
minimax version 1 is just a pixel shift modification to the existing VAE. Version 2 adds additional detail and also more accurately matches start-frame input images in detail, colour and lighting.
I will work on adding additional details to it if possible.
For minimax use this nodepacks vae decode:
https://github.com/TripleHeadedMonkey/ComfyUI-MiniMaxH3_LatentUpscaler
If the VAE still doesn't work for you (size mismatch error) then download the optional zipfile included and add that to "\ComfyUI\comfy\ldm\minimax\" replace the file vae.py" in this folder.
Description
Outputs an image 2x the input latent resolution with little resource cost.
FAQ
Comments (8)
Looks really amazing. I wish you would just share it and not paywall the damn thing.
Well, I apologise for gating it. But I want the Buzz to train H3 on Civitai, lol. It'll be worth it. :D
If you're nice and you DM me, I might just give it away to some poor people like myself :P
Not entirely sure why, but with this VAE, the official Comfy save image node throws an error saying it doesn't support this format, and the Pixaroma save image node only displays a black and white image. Is there anything specific I need to do?
(No problem with the regular VAE, of course.)
Yes there is a custom node "Vae decode (Vae utils)" Sorry I had forgotten to mention it.
I added a link to the pack in the description.
@Triple_Headed_Monkey Thanks, I’d come across that while doing some research, but I was waiting for your reply to be sure. It works extremely well. the buzz was well worth the spend. Thanks. :)
@Illynir I'm glad you like it :D