1 Sora LLM ChatGPT GPT-4 Sora Sora Sora Diffusion Transformer Sora Diffusion Transformer scale-up GPT-3 GPT-3 Token Transformer Sora OpenAI 32x 1x 4x 32x 1x 4x Sora Google Snap Diffusion Transformer Sora OpenAI-Sora Sora GPT-4V Diffusion Transformer LLM Sora DiT Diffusion 20 Steps patch LLM Sora 2 Sora.4 1.1 Sora.4.5 2.1 15 GAN.5 2.2 GAN VAE.6 2.3 Transformer Based.6 2.4 Diffusion Based.6 2.5 Diffusion Model.7 2.6 Diffusion.8 Sora.9 3.1 Video Encoding.9 3.2 Diffusion Transformer.11 3.3.12 3.4 Patch/Token.14 3.5.15 AI.16.17 5.1.18 5.2.18.19 1 Sora Scaling Law!2 Sora/.5 3.6 4 GAN.6 5 DiT Scaling Law.7 6 Genie.8 7.8 8 Sora.9 ZXFUuNtOtQnMpRpMqNtMoP7NbP6MoMmMmOtPfQmMtQlOmOmR7NoOzRNZqMtPNZpNnN 3 9 Sora Encoding.9 10 Patch.10 11 Sora.10 12 Navit.11 13 DiT.11 14 Sora.11 15.12 16.12 17.13 18.13 19.14 20 Diffusion.15 21/.15 22 LeCun.16 23 V-JEPA.16 24 AGI.16 25.17 26 AI.17 27 AI.18 28 AI-.18 29 Sora 3D.19 30 Genie.19 4 Sora 1.1 Sora Sora 60 Pika1.0 3s-7s Runway Gen-2 4s-18s Sora 1080P Minecraft Sora Scaling Law patch Sora 1 Sora Scaling Law Sora Technical Report Sora Sora Sora+BWqURRUU6TSZ/kb6W6pF87zkDSe1RnSOm2svFrHBRVmktmpBKUtlI6xMutIZRQEM 5 2 Sora/Sora Technical Report 2.1 15 GAN 1 2 3 4 6 3 Sora:A Review on Background 2.2 GAN VAE 2015 GAN VAE(Text2Filter TGANs-C)4 GAN Text-to-Video:The Task,Challenges and the Current State 2.3 Transformer Based(GPT-3)(DALL-E)Transformer Transformer Phenaki Make-A-Vide NUWA VideoGPT CogVideo Transformer TATS VQGAN Transformer Phenaki()NUWA-Infinity(autoregressive over autoregressive)2.4 Diffusion Based 7(3D)Video Diffusion Models(VDM)MagicVideo VDM Tune-a-Video-Video LDM Text2Video-Zero Runway Gen1 Runway Gen2 Stable Video Diffusion NUWA-XL 2.5 Diffusion Model Meta Sora William Peebles 23 3 Scalable Diffusion Models with Transformers Diffusion Transformers(DiTs)DiT Stable Diffusion UNet Transformer Scaling Law patch size 5 DiT Scaling Law Scalable Diffusion Models with Transformers Sora DiT 1 1080P Sora Sora Google Genie Spatiotemporal(ST)Transformers Stable Diffusion UNet Genie Genie 160X90 8 6 Genie Genie:Generative Interactive Environments Snap Spatiotemporal(ST)Transformers FIT Far-reaching Interleaved Transformers Token/Patch n Token/Patch Multi-head Self-Attention n2 FIT n4/3 2.6 Diffusion 15 6 9 Morph Studio 2-4 3-5 7/CapCut 4 4.4 3.5 16.1 Morph Studio 4 4 3.5 15.6 NeverEnds 3.8 3.8 2.9 14.4 AI 3.4 3 3.7 13.1 VideoCrafter 2 2.3 4 3.7 13 PixVerse 3.1 3 3.3 12.5 Vega AI 2.6 3 2.8 11.8 Pixeling 3.3 2.9 2.3 11.3 Video Diffusion Models Diffusion Transformer OpenSora Sora 3 CEO 9 Sora 8 Sora Sora:A Review on Background Sora Diffusion Transformer 1 Time-space compressor 2 ViT 3 CLIP LLM GPT-4 3.1 Video Encoding Sora 9 Sora Encoding Sora Technical Report Latent representation tokenization 2D 1D 16x 16 32x32 10 10 Patch Sora:A Review on Background Transformer Spacetime Patches Sora H*W patch patch HxWxT patch tube patch VIT embedding 3D patch t*H*W patch patch size HxWxT H W t patch patch patch Sora patch patch spacetime patch patch 11 Sora Sora Technical Report Sora ViT Vision Transformer patch Sora Navit Patch n Pack 11 3.2 Diffusion Transformer Sora Transformer Diffusion Model Scalable Diffusion Models with Transformers DiTs 12 Navit 13 DiT Sora:A Review on Background Sora:A Review on Background Stable Diffusion 1(Text Encoder)Transformer/Token 2(Image Information Creator)Stable Diffusion 3(Image Decoder)DiTs Sora U-Net Transformer Transformer Transformers Sora Transformer Sora Google Genie Snap Snap Videos ST-Transformer Spatio-temporal Transformer Sora Scaling Law Transformer GPT3.0 OpenAI 1x 4x 32x 32x 14 Sora Sora Technical Report 12 3.3 15 A Survey of Large Language Models 2024 3 1(Long-Context)Gemini1.5 Kimi Token 2(Multi-Modal)Token 3 MOE(Mixture-of-Experts)Mixtral Gemini1.5 Grok GPT MOE MOE 16 Opening AI Infrastructure by Meta prefill decode Token Token Token Loss Prefill Decode 13 17 18 Sequential Modeling for Reinforcement Learning Nvidia Prefill Phase key value prompt KV Cache Decoding Phase Token Token Token Token KV Cache Token embedding Token key value Prefill Phase KV Cache 14 3.4 Patch/Token 19 Opening AI Infrastructure by Meta Patch/Token Patch T W H Patch patch Ps Transformer Patch Multi-head Self-Attention Token Self-Attention Patch/Token 1 O(n2)2():O(n4)3 Patch O(n2)Sora 1080P 30/s 60 Patch 32x32 Patch 180 Token Gemini 1.5 Pro 100 Sora 20-50B GPT-4 Patch Prefill Decode 15 3.5 20 Diffusion How does Stable Diffusion work?Sora GPT Transformer Sora DiT Diffusion noise latent latent/20 Steps GPT Decoder-Only Transformer Auto Regression Token Sora DiT Encoder-Only Transformer Step Patch Patch GPT 21/(FP16)A100 624 TFLOPS 40/80GB HBM2e 2 TB/s H100 1979 TFLOPS 96GB HBM3 3.35 TB/s L40S 733 TFLOPS 48GB GDDR6 0.8 TB/s H20 148 TFLOPS 96GB HBM3 4 TB/s L20 119.5 TFLOPS 48GB GDDR6 0.8 TB/s TPU v5p 459 TFLOPS 95GB HBM3 2.76 TB/s 910B 376 TFLOPS 64GB HBM2e Groq 188 TFLOPS 230MB SRAM 80 TB/s SemiAnalysis Google TPU Arthurchiao H20 H100 LLM H100 10%H20 H100 Groq SRAM 80TB/s Token GPT Gemini LLM GDDR 16 AI 22 LeCun 23 V-JEPA A Path Towards Autonomous Machine Intelligence V-JEPA OpenAI Sora World Simulator World Model Meta Yann LeCun Meta Revisiting Feature Prediction for Learning Visual Representations from Video V-JEPA 24 AGI Meta OpenAI Scaling Law AGI AGI Sora AGI AGI Encoder 17 Yann LeCun Llama3 GPT-5 AGI 25 Sora:A Review on Background CapCut Krea Runway Sora 26 AI SimilarWeb 18 5.1 Sora AI 24 2 23 AI AI AI 1 3 29 AI-AI Stable Diffusion Sora 27 AI 28 AI-CCTV 5.2 1 2 3 Sora 3D 3D prompt 19 29 Sora 3D 30 Genie Sora Technical Report Genie:Generative Interactive Environments 4 Google Genie 1.Transformer 2.3.AI AI 20 C3(C3 021-80234211 201204 1088 5 010-85950438 100005 26 8 0755-86695353 518000 2028 18 1806