We believe that the future image generation paradigm should be more simple and flexible, that is, generating various images directly through arbitrarily multi-modal instructions without the need for additional plugins and operations, similar to how GPT works in language generation.

Existing image generation models often need to load multiple additional network modules (such as ControlNet, IP-Adapter, Reference-Net, etc.) and perform additional preprocessing steps (such as face detection, pose estimation, cropping, etc.) to generate satisfactory images. But we believe that the future image generation paradigm should be more simple and flexible, that is, generating various images directly through arbitrarily multi-modal instructions without the need for additional plugins and operations, similar to how GPT works in language generation.