创意人工智能研究组 AI for Creativity Research Team

» 从“素材生成器”走向“具有设计思维的创意智能体”。From asset generator to design-thinking creative agent.

我们关注 AI 如何获得在创作中进行顶层设计的能力。导演构思叙事作品时，本质上是在脑海中同时组织声音、画面、动作、节奏与故事逻辑，进行跨模态的整体想象。本方向重点研究多模态感知、跨模态语义对齐、设计元素建模、美学认知与创作推理，使模型能够理解声音、画面、叙事与情绪之间的复杂关系。A film director, when conceiving a narrative, orchestrates sound, image, action, pacing, and story logic simultaneously — a cross-modal act of imagination. This pathway studies how AI can acquire that top-level design capability through multimodal perception, cross-modal alignment, aesthetic cognition, and creative reasoning.

• 多模态感知Multimodal Perception
• 跨模态对齐Cross-modal Alignment
• 创作推理Creative Reasoning
• 美学认知Aesthetic Cognition
• 设计思维智能体Design-Thinking Agent

» 让工具适应创作，而不是让艺术家适应工具。Let the tools adapt to creators — not the other way around.

我们关注如何将创意智能体真正引入内容生产流程，减少 2D / 3D 动画制作中视觉预览、冷启动建模、补间帧、多镜头衔接等耗时环节的重复劳动。覆盖从 1D 音频、2D 图像、3D 几何资产到 4D 视频的全数据形态，打破单一图文驱动范式，构建更贴近影视动画创作直觉的多模态交互机制。We bring creative agents into real production, reducing mechanical steps in 2D / 3D animation: previsualization, cold-start 3D modeling, inbetweening, shot transitions, asset recomposition. We cover the full data spectrum — 1D audio, 2D images, 3D geometry, 4D video — and break the single text-to-image paradigm with multimodal interaction grounded in artist intuition.

• 智能体落地Agent Deployment
• 2D/3D/4D 内容2D/3D/4D Content
• 多模态交互Multimodal Interaction
• 影视动画工作流Film & Animation Workflow
• Workflow → Creative FlowWorkflow → Creative Flow