search

PixArt-alpha / PixArt-sigma

PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation
https://pixart-alpha.github.io/PixArt-sigma-project/
GNU Affero General Public License v3.0
1.44k stars 68 forks source link

VAE for the DMD model #56

Closed EternalEvan closed 2 months ago

EternalEvan commented 2 months ago

Hello, I noticed that the pre-trained pipeline for DMD model has a different VAE compared with the PixART-Sigma-MS-1024 which uses SDXL's VAE. And I try to use the SDXL's VAE for DMD model but it gives bad results. So what's the VAE for the DMD model? Have you tuned the VAE yourself?

lawrence-cj commented 2 months ago

It's just the PixArt-Alpha's VAE, also for the SDV1, V2 and so on.