Support for Qwen-Image

[zheyuanliu233]

Hi, great work on the project!

I noticed the current vision encoder is based on FLUX. The recently released Qwen-Image model seems to offer significantly better performance with a similar architecture.

Are there any plans to adapt the project to use Qwen-Image?

Thanks!

[mi804]

Qwen-Image itself uses qwen2.5 vl-7b as the text encoder, so we won't repeat the same process on Qwen-Image. However, we will continue to follow and explore directions towards unified understanding and generation, and may further optimize the Nexus-Gen architecture in the future.