Tag
Someone compared GPT-6.2 Astra and Sol across tasks including 3D scenes, games, slideshows, data visualization, websites, and interactive apps, highlighting Astra's advantages in detail and precision alongside Sol's strengths in motion design, with the cost of each run during the recording included.
The paper proposes Variable Length Diffuse Everything (VL-DE), a general framework for variable-length multimodal generation on arbitrary state spaces using an insertion-based mechanism instead of padding, along with a higher-order sampler, demonstrating effectiveness on layout generation.
DySENSE is a joint multimodal GenAI framework that synthesizes static urban satellite imagery alongside spatially aligned, dynamic monthly building energy consumption maps, decoupling long-term urban morphology from short-term energy dynamics. Evaluated on four cities, it achieves strong downstream performance with less than 25% labeled data.
本文提出四层级多模态生成模型能力框架,并推出 WorldCast 基准(320 个案例、8 类任务),评测 12 个代表性模型的多步世界状态预测能力,发现前沿模型(如 GPT-Image-2.5)进展迅速,而开源模型在推理密集型视觉预测上差距明显。
Intern Lumina U2 is presented as the first diffusion language model using multi-codebook discrete visual tokens to unify image, video, and 3D generation with text in a single backbone, achieving leading performance and 4.45x faster image generation via stability-aware sampling.
This paper introduces LEDFlow, an entropy-guided sampler for uniform discrete flow that improves generation accuracy by adaptively ordering predictions based on local entropy, with gains on reasoning, image generation, and multimodal tasks.
SGLang-Diffusion with VDN-H3 from MiniMax enables fast and scalable video generation, achieving 14.4s of 768p video in just 9.0s on 8× B200 GPUs for faster-than-real-time inference.
GenFirst proposes a generation-before-reconstruction strategy to enable stable end-to-end training of latent generative models, avoiding latent collapse and achieving state-of-the-art results in image synthesis and multimodal generation.
This paper introduces G-CARL, a grounded checklist-aligned reinforcement learning framework for patient-oriented medical report interpretation, along with the MMedReport benchmark, demonstrating improved factuality and alignment with patient needs.
SPARGen introduces a unified multimodal framework that casts 3D reconstruction, dense correspondence, and spatial reasoning as instruction-conditioned generation tasks, enabling shared spatial representations.
Introduces MCTS-Report, a Monte Carlo Tree Search framework for generating multimodal reports from tabular data, along with the MMRBench benchmark. It outperforms strong baselines across structural completeness, numerical accuracy, chart-text alignment, and insight novelty.
This paper presents SenseNova-Vision, a unified multimodal model that formulates computer vision tasks as generation problems, achieving performance comparable to specialized systems across diverse vision tasks. It introduces a large-scale instruction-response corpus and publicly releases the model and datasets.
PairCoder++ introduces a pair programming paradigm where a Driver agent writes code and a Navigator agent reviews it using toolchain verification, achieving significant improvements across 17 benchmarks for generating charts, scientific figures, CAD models, 3D scenes, and other structured artifacts.
This paper proposes NaviGen, a framework for personalized multimodal content generation that encodes user behavior into executable instructions using a dual identifier and a two-stage SFT+RL pipeline, improving personalization across product, game, and short-video domains.
TIGER is an inference-time framework that mitigates hallucinations in multimodal generation by extracting observation and claim graphs and assigning risk scores to repair unsupported facts. It reduces unsupported content across image-to-text, image+text-to-text, audio-to-text, and video-to-text tasks.
AlphaGRPO is a new framework that applies Group Relative Policy Optimization to Unified Multimodal Models, enhancing generation through self-reflective refinement and decompositional verifiable rewards.
STARFlow2 is a new research paper introducing an architecture that bridges language models and autoregressive normalizing flows for unified multimodal generation. It addresses structural mismatches in existing systems by using a shared causal masking mechanism for interleaved text-image sequences.
MM-WebAgent is a hierarchical agentic framework that generates coherent and visually consistent webpages by coordinating AIGC-based element generation through joint optimization of layout and multimodal content. The paper introduces a benchmark and multi-level evaluation protocol, demonstrating improvements over code-generation and agent-based baselines.