创意人工智能研究院 AIGC Research

我们旨在打破传统 AI“只能机械生成、无法理解创作”的局限，研究人类艺术创作的意图、过程与认知机制，赋予AI类似人类导演或设计师的“设计思维 (Design Thinking)”，并借鉴神经科学探索下一代创意智能架构。

1. 跨模态美学对齐：优秀的艺术作品源于对声音、画面与故事的整体想象。我们致力于文字、音乐、色彩、情感与动态效果等设计元素的深度对齐，赋予 AI 美学感知、创作推理与全景构思能力。
2. 大模型的可解释性与轻量化：从微观（神经元激活表征）、中观（子网络与注意力机制协同）到宏观（整体泛化与涌现能力）三个尺度，深度剖析生成式大模型的内在机理。通过揭示这些“黑盒”规律，指导算法优化与架构精简，实现“以小博大”的高效、轻量化计算。
3. 类脑启发与科学赋能：向内借鉴 (仿真驱动的类脑研究)：将神经机制与脑网络转化为可计算模型，探索下一代智能架构；向外赋能 (AI驱动的神经科学)：同时运用生成式 AI 与数据分析反哺神经科学研究，辅助探索神经连接及疾病机理研究，发掘并解决具有重大价值的科学难题，促进 AI 与脑科学双向发展。

We aim to move beyond traditional AI's limitation of mechanically generating content without understanding creativity. We study the intentions, processes, and cognitive mechanisms behind human artistic creation, equip AI with design thinking akin to that of human directors or designers, and draw on neuroscience to explore next-generation creative intelligence architectures.

1. Cross-modal Aesthetic Alignment: Outstanding artworks emerge from a holistic vision of sound, visuals, and story. We pursue deep alignment across design elements such as text, music, color, emotion, and motion, enabling AI to perceive aesthetics, reason creatively, and conceive works as a whole.
2. Interpretability and Lightweighting of Large Models: We analyze generative models at three scales: micro (neuron activation patterns), meso (coordination among subnetworks and attention mechanisms), and macro (overall generalization and emergent capabilities). By revealing how these black boxes work, we guide algorithmic optimization and architectural refinement to achieve efficient, lightweight computation that does more with less.
3. Brain-inspired Research and Scientific Empowerment: Inward inspiration (simulation-driven brain-inspired research) turns neural mechanisms and brain networks into computable models to explore next-generation intelligent architectures. Outward empowerment (AI-driven neuroscience) uses generative AI and data analysis to advance neuroscience, help investigate neural connectivity and disease mechanisms, identify and solve significant scientific challenges, and promote progress in both AI and brain science.

我们聚焦影像叙事与影视创作，将前沿 AI 能力转化为创作者可用的工具，减少动画和影视制作中的重复劳动，让创作者更专注于故事与表达。

1. 全维度多模态交互：研究从音频、原画、三维资产到时空视频的内容形态，探索动作、音频与多模态协同驱动的创作方式。
2. 影像叙事与镜头语言：围绕长视频下的角色一致性、动作连续性、剧情对齐与镜头语言，研究能够服务完整故事表达的智能影像方法。
3. 创作流程落地：重新思考并推动生成式 AI 融入影视制作流程，发掘当下创作流中的"灯下黑"问题，开创AI原生的后期制作新形态，从繁琐工作流（Workflow）走向自由创意流（Creative Flow）。

We focus on visual storytelling and filmmaking, turning cutting-edge AI capabilities into tools creators can use and reducing repetitive work in animation and film production so creators can focus on story and expression.

1. Full-spectrum Multimodal Interaction: We study content forms ranging from audio and concept art to 3D assets and spatiotemporal video, exploring creation driven by motion, audio, and multimodal collaboration.
2. Visual Storytelling and Cinematic Language: For long-form video, we study character consistency, action continuity, plot alignment, and cinematic language to develop intelligent visual methods that support complete storytelling.
3. Creative Workflow in Practice: We rethink and advance the integration of generative AI into filmmaking, uncover overlooked problems in today's creative process, and pioneer AI-native forms of post-production, moving from cumbersome workflows (Workflow) toward free creative flow (Creative Flow).

星联无限灵感工作室连接创意智能研究与真实创作实践，探索人工智能技术与影像、故事及视觉表达的融合，让灵感从构思走向作品，并将创作中的经验反馈给研究。

1. 灵感共创：围绕故事构思、视觉设计与跨媒介表达，探索人机协同创作的发散、收敛与迭代。
2. 作品实践：结合实验室研究与创作者经验，制作并迭代可展示的影像作品与创意原型。
3. 创作反馈：从真实创作过程提炼需求与问题，推动智能创作工具和影像叙事方法持续改进。

StarX Infinite Inspiration Studio connects creative intelligence research with real creative practice. We explore how AI can work with visual storytelling and expression, helping ideas become tangible works and bringing lessons from making back into research.

1. Co-creating Ideas: Across story concepts, visual design, and cross-media expression, we explore divergence, convergence, and iteration in human–AI co-creation.
2. Creative Practice: We combine laboratory research with creators' experience to make and refine visual works and creative prototypes that can be shared.
3. Learning from Creation: We draw needs and questions from real creative work to keep improving intelligent tools and visual storytelling methods.