Search papers, labs, and topics across Lattice.
UniCom is introduced as a unified multimodal framework that bridges understanding and generation by compressing continuous semantic representations, addressing the limitations of discrete visual tokenizers. The approach uses an attention-based semantic compressor to distill dense visual features into a compact unified representation, finding that reducing channel dimension is more effective than spatial downsampling. Experiments show UniCom achieves SOTA generation performance among unified models, with improved controllability in image editing and consistency.
Ditch discrete visual tokens: UniCom achieves SOTA multimodal generation by compressing continuous semantic representations, unlocking better controllability and consistency in image editing.
Current unified multimodal models typically rely on discrete visual tokenizers to bridge the modality gap. However, discretization inevitably discards fine-grained semantic information, leading to suboptimal performance in visual understanding tasks. Conversely, directly modeling continuous semantic representations (e.g., CLIP, SigLIP) poses significant challenges in high-dimensional generative modeling, resulting in slow convergence and training instability. To resolve this dilemma, we introduce UniCom, a unified framework that harmonizes multimodal understanding and generation via compressed continuous representation. We empirically demonstrate that reducing channel dimension is significantly more effective than spatial downsampling for both reconstruction and generation. Accordingly, we design an attention-based semantic compressor to distill dense features into a compact unified representation. Furthermore, we validate that the transfusion architecture surpasses query-based designs in convergence and consistency. Experiments demonstrate that UniCom achieves state-of-the-art generation performance among unified models. Notably, by preserving rich semantic priors, it delivers exceptional controllability in image editing and maintains image consistency even without relying on VAE.
