Unverified50% confidenceFactExact time
图像嵌入向量的维度通常在 512 到 2048 之间
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
变长矩阵分类：如何为无序集合数据选择神经网络架构
Related Claims
UnverifiedThe native resolution of mainstream image generation models (such as Stable Diffusion and DALL-E) typically ranges from 512×512 to 1024×1024.73% similarUnverified一张图片可能产生256到576个token69% similarUnverifiedCLIP将文本和图像映射到同一个高维向量空间中，通常是512维或768维69% similarUnverifiedImagen系列自2022年首次发布，采用级联扩散架构实现从64×64到1024×1024的逐级超分辨率生成67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/558827API
curl https://kongchang.com/api/v1/knowledge/claims/558827MCP
get_claim(id=558827)