bagel-image-t2i

BAGEL is a multi-modal unified model open-sourced by ByteDance's Seed team, adopting the MoT architecture with 7B activated parameters (totaling 14B). It is pre-trained on large-scale interleaved text/image/video/webpage data, capable of visual understanding, text-to-image generation, and image editing. It supports conversational usage, automatically determining whether to understand or generate, providing an end-to-end integrated experience.