Draw-In-Mind: Rebalancing Designer-Painter Roles in Unified Multimodal Models Benefits Image Editing

Researchers introduce Draw-In-Mind (DIM), a new approach to multimodal AI models that improves image editing by better balancing responsibilities between understanding and generation modules. The DIM-4.6B model achieves state-of-the-art performance on image editing benchmarks despite having fewer parameters than competing models.

• →DIM addresses limitations in unified multimodal models by rebalancing designer-painter roles between understanding and generation modules.
• →The approach uses a dataset with 14M long-context image-text pairs and 233K chain-of-thought imaginations from GPT-4o.
• →DIM-4.6B-Edit achieves SOTA performance on ImgEdit and GEdit-Bench benchmarks with modest parameter scale.
• →The model outperforms much larger models like UniWorld-V1 and Step1X-Edit in image editing tasks.
• →Explicitly assigning design responsibility to the understanding module provides significant benefits for image editing accuracy.