Search papers, labs, and topics across Lattice.
This paper introduces DeltaV, a Unified Large Multimodal Model (ULMM) that enhances multimodal reasoning by replacing full-image generation with incremental visual updates, significantly reducing visual-token redundancy. By employing a temporal similarity (TSIM) Router, DeltaV efficiently allocates tokens based on the magnitude of visual changes, leading to a 55.6% reduction in newly generated visual tokens while maintaining reconstruction fidelity. Experimental results demonstrate that DeltaV improves multimodal reasoning performance by 3.3% compared to traditional full-image generation and outperforms larger models on various benchmarks.
Visual updates in DeltaV cut token generation by over half while boosting reasoning accuracy, challenging the need for full-image outputs in multimodal models.
Current Unified Large Multimodal Models (ULMMs) support interleaved multimodal reasoning through textual reasoning and intermediate visual states, but typically generate each visual state as a full image. This full-image generation paradigm introduces substantial visual-token redundancy and dilutes supervision on sparse yet reasoning-critical state transitions. We propose DeltaV, a ULMM that replaces full-image generation with visual updates. Conditioned on historical visual states, DeltaV incrementally predicts compact update tokens that capture the visual changes across reasoning steps, avoiding repeated modeling of unchanged content. To align the token budget of each update with the magnitude of visual change, DeltaV introduces a temporal similarity (TSIM) Router, which stops allocating tokens once the marginal reconstruction gain falls below a threshold. To support more diverse and generalizable reasoning, we further construct StructCoT, a large-scale interleaved multimodal reasoning dataset with 1.05M samples spanning 44 task domains. Experiments show that the visual-update paradigm reduces newly generated visual tokens by 55.6\% on average without compromising reconstruction fidelity, and improves multimodal reasoning by 3.3\% over full-image generation. Trained with StructCoT and large-scale multimodal data, DeltaV-2B further outperforms substantially larger open-source models by 8.4\% on in-domain multimodal reasoning evaluations and surpasses the comparable-scale Qwen3-VL-2B by 5.9\% on external multimodal reasoning and understanding benchmarks. Code, models, and StructCoT will be released at https://github.com/Pengjie-W/DeltaV.
