Search papers, labs, and topics across Lattice.
DeepGen 1.0, a 5B parameter unified multimodal model, is introduced to address the high computational costs associated with large-scale image generation and editing models. The model employs Stacked Channel Bridging (SCB) to extract and fuse hierarchical features from multiple Vision Language Model (VLM) layers, guiding the generative backbone with structured reasoning. A data-centric training strategy involving alignment pre-training, joint supervised fine-tuning, and reinforcement learning with MR-GRPO enables DeepGen 1.0 to achieve state-of-the-art performance on various benchmarks, surpassing models like HunyuanImage and Qwen-Image-Edit despite being trained on a relatively small dataset.
A 5B model just crushed the image generation and editing performance of models 5-16x larger, thanks to smarter feature fusion and a novel RL training strategy.
Current unified multimodal models for image generation and editing typically rely on massive parameter scales (e.g.,>10B), entailing prohibitive training costs and deployment footprints. In this work, we present DeepGen 1.0, a lightweight 5B unified model that achieves comprehensive capabilities competitive with or surpassing much larger counterparts. To overcome the limitations of compact models in semantic understanding and fine-grained control, we introduce Stacked Channel Bridging (SCB), a deep alignment framework that extracts hierarchical features from multiple VLM layers and fuses them with learnable'think tokens'to provide the generative backbone with structured, reasoning-rich guidance. We further design a data-centric training strategy spanning three progressive stages: (1) Alignment Pre-training on large-scale image-text pairs and editing triplets to synchronize VLM and DiT representations, (2) Joint Supervised Fine-tuning on a high-quality mixture of generation, editing, and reasoning tasks to foster omni-capabilities, and (3) Reinforcement Learning with MR-GRPO, which leverages a mixture of reward functions and supervision signals, resulting in substantial gains in generation quality and alignment with human preferences, while maintaining stable training progress and avoiding visual artifacts. Despite being trained on only ~50M samples, DeepGen 1.0 achieves leading performance across diverse benchmarks, surpassing the 80B HunyuanImage by 28% on WISE and the 27B Qwen-Image-Edit by 37% on UniREditBench. By open-sourcing our training code, weights, and datasets, we provide an efficient, high-performance alternative to democratize unified multimodal research.
