Unverified50% confidenceFactExact time
扩散变换器（DiT）架构最早由Meta研究院在2022年提出，将扩散模型的噪声预测网络替换为Transformer架构，Sora、Kling、Wan等主流视频生成模型均采用此路线
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedThe DiT architecture was formally proposed by Meta Research in 2023, introducing the Vision Transformer into the denoising process of diffusion models, with computational complexity growing quadratically with sequence length.83% similarVerifiedDiT was proposed by Peebles and Xie in 2023; it replaces the U-Net backbone in traditional diffusion models with Transformer blocks and has become the technical foundation for video generation models such as Sora.78% similarVerifiedThe Diffusion Transformer (DiT) architecture was first proposed by Meta Research in 2022.77% similarUnverifiedDiT was proposed by Peebles and Xie in 2022, introducing the Transformer architecture into the diffusion process.73% similarUnverifiedDify是2023年兴起的开源LLMOps（大模型运维）平台67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/518242API
curl https://kongchang.com/api/v1/knowledge/claims/518242MCP
get_claim(id=518242)