Unverified50% confidenceFactExact time
在Stable Diffusion中每个token被映射为768维或1024维的嵌入向量
1
Sources
50%
Confidence
Long-term
Relevance
8/20/2026
First Seen
Sources
Related Claims
UnverifiedStable Diffusion 3.5的潜在空间输出约65000维（16×64×64）75% similarUnverifiedStable Diffusion XL's model weights alone occupy approximately 6.5 GB of VRAM, and generating a single 1024×1024 image typically requires more than 10 GB of GPU VRAM.74% similarUnverified在1024维空间中要求向量完全正交最多只能塞进1024个向量，但允许近似正交（夹角81.4°到98.6°之间）时可容纳超过10万个向量67% similarUnverifiedStable Diffusion compresses a 512×512×3 image into a 64×64×4 latent representation via VAE, achieving a compression ratio of 48×.64% similarUnverifiedTo run Stable Diffusion XL base version, an NVIDIA RTX 3060 (12GB VRAM) can perform inference smoothly; the full version of Flux.1-dev (FP16 precision) requires approximately 24GB of VRAM.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/777418API
curl https://kongchang.com/api/v1/knowledge/claims/777418MCP
get_claim(id=777418)