Version Next. Using EQVAE trained with 1.6k dataset repeated 8 times. This is a proof of concept, and some results will be having issues with color since the latent space is not trained enough.

[HIGHLY NOT COMPATIBLE WITH SDXL VAE, USE BUILT-IN VAE!]

If you want to see the end product, you can contact me for giving sponsorship. The end product only needs at least 250k dataset by initial calculation.

Trained from: Animagine XL 4.0-Zero

@kblueleaf's EQVAE: https://huggingface.co/KBlueLeaf/EQ-SDXL-VAE

Using the same license as Animagine XL 4.0 and UrangDiffusion.