Organizations: Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China · Beijing Key Laboratory of Research on Large Models and Intelligent Governance · Engineering Research Center of Next-Generation Intelligent Search and Recommendation, MOE · Ant Group
We present LLaDA-o, an effective and length-adaptive omni diffusion model for multimodal understanding and generation. LLaDA-o is built on a Mixture of Diffusion (MoD) framework that decouples discrete masked diffusion for text understanding and continuous diffusion for visual generation, while coupling them through a shared, simple, and efficient attention backbone that reduces redundant computation for fixed conditions. Building on MoD, we further introduce a data-centric length adaptation strategy that enables flexible-length decoding in multimodal settings without architectural changes. Extensive experiments show that LLaDA-o achieves state-of-the-art performance among omni-diffusion models on multimodal understanding and generation benchmarks, and reaches 87.04 on DPG-Bench for text-to-image generation, supporting the effectiveness of unified omni diffusion modeling. Code is available at https://github.com/ML-GSAI/LLaDA-o.
Figures & tables
Figure 1: Overview of LLaDA-o’s capabilities. Top: multimodal understanding examples. Middle: text-to-image generation results following complex prompts (see Table 7 for the prompts). Bottom: case by case comparison with existing omni diffusion models, where LLaDA-o achieves stronger understanding performance and generates images with richer fine-grained details following the instructions.
Figure 2: Overview of LLaDA-o: the mixture of diffusion framework.
Figure 3: Implementation of intra-modality bidirectional attention. Yellow blocks indicate unmasked attention, while dashed white boxes denote masked attention. Text sequences are explicitly partitioned into Prompts ( PRM ) and Responses ( RES ) in cases (a–b).
Figure 4: Comparison of inference efficiency on MathVista. We visualize the throughput-accuracy trade-off by varying the confidence threshold for LLaDA-o and the refresh interval ( n ) of Fast-dLLM applied to LLaDA-V. Our approach outperforms LLaDA-V, achieving a 5.9× speedup with comparable performance.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 5: Additional generated samples. We present 12 randomly selected images generated by LLaDA-o. For each sample, the prompt used for generation is shown below the corresponding image. All results are produced under the same setting as in the main paper.
Table 9: Qualitative samples demonstrating variable-length generation. Top: Under a long block length setting ( L=64 ), LLaDA-V generates redundant content for a simple prompt, while LLaDA-o correctly terminates. Bottom: Under a short block length setting ( L=16 ), LLaDA-V produces an overly brief response for a detailed prompt, whereas LLaDA-o dynamically adapts the output length to the user prompt.
Table 10: Qualitative samples demonstrating variable-length generation. We show generations from LLaDA-o for the same image and the same prompt under different block length settings ( L∈{16,32,64,128} ). Despite changing L , the output length remains largely stable, indicating that LLaDA-o can dynamically adapt the output length to the user prompt and image.
We present LLaDA2.0-Uni, a unified discrete diffusion large language model (dLLM) that supports multimodal understanding and generation within a natively integrated framework. Its architecture combines a fully semantic discrete tokenizer, a MoE-based dLLM backbone, and a diffusion decoder. By discretizing continuous visual inputs via SigLIP-VQ, the model enables block-level masked diffusion for both text and vision inputs within the backbone, while the decoder reconstructs visual tokens into high-fidelity images. Inference efficiency is enhanced beyond parallel decoding through prefix-aware optimizations in the backbone and few-step distillation in the decoder. Supported by carefully curated large-scale data and a tailored multi-stage training pipeline, LLaDA2.0-Uni matches specialized VLMs in multimodal understanding while delivering strong performance in image generation and editing. Its native support for interleaved generation and reasoning establishes a promising and scalable paradigm for next-generation unified foundation models. Codes and models are available at https://github.com/inclusionAI/LLaDA2.0-Uni.
Unified multimodal diffusion large language models (dLLMs) offer a single architecture for both image generation and multimodal understanding, but their iterative decoding requires tens to hundreds of forward passes. Existing few-step distillation methods largely focus on either image generation or text generation, making it unclear how to compress a fully discrete multimodal dLLM into a single efficient student while preserving both generation and understanding. We introduce Omni-Diffusion-Distill, a unified two-stage distillation framework that retains strong generation and understanding capabilities while substantially reducing the inference cost of a unified multimodal dLLM. Omni-Diffusion-Distill aligns the distillation of both generation and understanding, for both images and text, in the discrete token space. In the first stage, the student is trained to skip decoding steps by replaying cached teacher trajectories, and in the second stage the student is refined on intermediate states along its own rollouts. We further remedy two sources of degradation in unified distillation with a pairwise collision penalty that reduces repetition under parallel text decoding, and entropy-matched guidance that prevents entropy collapse caused by fitting the sharpened teacher distribution in image generation. Omni-Diffusion-Distill achieves state-of-the-art trade-offs between decoding efficiency and generation and understanding performance for multimodal dLLMs, reducing image generation from 128 to 8 decoding steps and multimodal understanding from 512 to 64, giving 18.2x and 21.2x wall-clock speedups. Under these budgets, it scores 0.828 on GenEval and 83.0 on DPG-Bench for text-to-image generation, while reaching GPT judge scores of 20.0 on MM-Vet and 57.2 on COCO captioning (twice the teacher's 28.4 at the same steps) for multimodal understanding.
While recent multimodal large language models (MLLMs) have made impressive strides, they predominantly employ a conventional autoregressive architecture as their backbone, leaving significant room to explore effective and efficient alternatives in architectural design. Concurrently, recent studies have successfully applied discrete diffusion models to various domains, such as visual understanding and image generation, revealing their considerable potential as a promising backbone for multimodal systems. Drawing inspiration from these pioneering studies, we introduce Omni-Diffusion, the first any-to-any multimodal language model built entirely on mask-based discrete diffusion models, which unifies understanding and generation across text, speech, and images. Omni-Diffusion employs a unified mask-based discrete diffusion model to directly capture the joint distribution over discrete multimodal tokens. This approach supports not only bimodal tasks but also more complex scenarios involving multiple modalities. On a diverse set of benchmarks, our method outperforms or performs on par with existing multimodal systems that process two or more modalities, highlighting the significant promise of diffusion models in powering the next generation of multimodal foundation models. Project webpage: https://omni-diffusion.github.io.