通往多模态通用人工智能之路.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Jason LevinDavid Barona(Davicho)The path to Multimodal Artificial General IntelligenceS T P 1 0 2 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.

2、Language models understand textAGI must understand the physical worldINDUSTRY FIRST REASONING VIDEO MODEL AND THE FIRST TO GENERATE HIGH-FIDELITY 16-BIT HDRMODEL STRENGTHSPhysics Simulations,Particles,Fluids and ExplosionsRay3 delivers high motion fidelity with stable simulations,natural particle fl

3、ow,coherent fluids,and consistent explosive dynamics.World Motion,Understanding&ExplorationRay3 shows strong spatial reasoning,maintaining depth,perspective,and seamless motion with stable scale and coherence across environments.Character Anatomy,Action&AnimationRay3 preserves anatomical accuracy,ex

4、pressive movement,and fluid action,keeping characters consistent,believable,and impactful.Closeup Details&Lighting Effects Ray3 preserves fine detail with clear textures,surfaces,and lighting,handling reflections,glow,and shadows while staying consistent.NEW FEATURESRay3 can export your creations in

5、 High Dynamic Range,delivering the precision required for VFX,compositing,and post-production,including support for 10,12 and 16 bit EXR files to preserve every layer of detail and flexibility for professional workflows.HDR+EXRSDRRaw Flat HDRNightDayNew FeaturesReasoningRay3 Reasoning judges and ref

6、ines outputs in real time for stable,consistent results.New FeaturesAnnotationRay3 supports Visual Annotation,letting creators scribble directly onto frames to control motion,shape actions,and add new visual elements.New FeaturesRay3s Draft Mode enables low-res 360p previews before full renders,savi