DreamLLM is a powerful multimodal language model learning framework. It achieves a synergy between multimodal understanding and creation. This open-source tool provides core functionalities such as multimodal understanding, raw multimodal space sampling, and interleaved document generation. DreamLLM performs excellently in zero-shot scenarios and is suitable for various multimodal tasks and applications. With a special dream token, it can predict image generation locations, providing users with powerful image generation capabilities.