Unverified50% confidenceFactExact time
Krea 2模型在底层架构上融合了扩散Transformer（DiT）思路，优于早期的U-Net架构模型
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedDiT 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66% similarUnverified图像生成领域正在经历从U-Net到Transformer的架构范式转移，传统扩散模型如Stable Diffusion 1.x和2.x主要依赖U-Net编码器-解码器结构64% similarVerifiedFLUX采用了DiT（Diffusion Transformer）架构，用纯Transformer替代了传统的U-Net结构64% similarUnverifiedL2约简成为后续fplll等高性能格约简库的理论基石63% similarUnverifiedBDI模型随后被Rao和Georgeff形式化为可计算的智能体架构，成为早期多智能体系统如JADE平台的理论基础62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/512048API
curl https://kongchang.com/api/v1/knowledge/claims/512048MCP
get_claim(id=512048)