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.

现有的图像生成模型往往需要加载多个额外的网络模块（如 ControlNet、IP-Adapter、Reference-Net 等）并执行额外的预处理步骤（例如人脸检测、姿势估计、裁剪等）才能生成令人满意的图像。但我们认为未来的图像生成范式应该更加简单灵活，即直接通过任意多模态指令生成各种图像，而无需额外的插件和操作，类似于 GPT 在语言生成中的工作方式。