Verified80% confidenceFactTime unknown
FLUX采用了DiT（Diffusion Transformer）架构，用纯Transformer替代了传统的U-Net结构
8
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
SVDQuant：4-bit量化让扩散模型在消费级GPU上高效运行
githubnunchaku-ai
Related Entities
Related Claims
VerifiedLTX 2.3采用基于Transformer的扩散架构（DiT，Diffusion Transformer），相较于传统U-Net架构具有更强的时序一致性建模能力78% similarPartially VerifiedSora采用DiT（Diffusion Transformer）架构，将视频压缩为时空patch序列后输入Transformer，以Transformer替代传统U-Net作为去噪骨干网络77% similarUnverifiedDiT 架构用 Transformer 替代传统 U-Net 作为扩散模型骨干网络，能在 Patch 级别捕捉跨帧长程依赖关系76% similarUnverifiedWan 2.2 属于扩散变换器（Diffusion Transformer，DiT）架构76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/28968API
curl https://kongchang.com/api/v1/knowledge/claims/28968MCP
get_claim(id=28968)