Paper Detail
Zijie Li, Yichun Shi, Jingxiang Sun, Ye Wang, Yixuan Huang, Zhiyao Guo, Xiaochen Lian, Peihao Zhu, Yu Tian, Zhonghua Zhai, Peng Wang
We present MMCORE, a unified framework designed for multimodal image generation and editing. MMCORE leverages a pre-trained Vision-Language Model (VLM) to predict semantic visual embeddings via learnable query tokens, which subsequently serve as conditioning signals for a diffusion model. This streamlined design effectively transfers the rich understanding and reasoning capabilities of VLMs into the visual generation process. By obviating the need for deep fusion between autoregressive and diffusion models or training from scratch, MMCORE significantly reduces computational overhead while maintaining high-fidelity synthesis. MMCORE seamlessly integrates text-to-image synthesis with interleaved image generation, demonstrating robust multimodal comprehension in complex scenarios such as spatial reasoning and visual grounding. Comprehensive evaluations indicate that MMCORE consistently outperforms state-of-the-art baselines across a broad spectrum of text-to-image and single/multi-image editing benchmarks.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{li2026mmcore,
title = {MMCORE: MultiModal COnnection with Representation Aligned Latent Embeddings},
author = {Zijie Li and Yichun Shi and Jingxiang Sun and Ye Wang and Yixuan Huang and Zhiyao Guo and Xiaochen Lian and Peihao Zhu and Yu Tian and Zhonghua Zhai and Peng Wang},
year = {2026},
abstract = {We present MMCORE, a unified framework designed for multimodal image generation and editing. MMCORE leverages a pre-trained Vision-Language Model (VLM) to predict semantic visual embeddings via learnable query tokens, which subsequently serve as conditioning signals for a diffusion model. This streamlined design effectively transfers the rich understanding and reasoning capabilities of VLMs into the visual generation process. By obviating the need for deep fusion between autoregressive and diffu},
url = {https://huggingface.co/papers/2604.19902},
keywords = {Vision-Language Model, diffusion model, semantic visual embeddings, learnable query tokens, text-to-image synthesis, image generation, visual grounding, multimodal comprehension, high-fidelity synthesis, huggingface daily},
eprint = {2604.19902},
archiveprefix = {arXiv},
}
{}