Unverified50% confidenceFactExact time
Mainstream diffusion models are typically trained at specific resolutions (such as 512×512 or 1024×1024), and extrapolation failures occur when generating high-resolution images that exceed the training distribution.
1
Sources
50%
Confidence
Long-term
Relevance
9/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe native resolution of mainstream image generation models (such as Stable Diffusion and DALL-E) typically ranges from 512×512 to 1024×1024.77% similarUnverifiedGenerating a single 512×512 resolution image with a diffusion model typically requires 20–50 denoising iterations, with a single-image generation latency of approximately 2–8 seconds on consumer-grade GPUs.71% 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68% similarUnverified经典CLIP在相对较低的分辨率（如224×224）上训练，对于密集文本、小目标或高分辨率图表信息会被严重压缩65% similarUnverifiedThe frame decoder uses 4-level bilinear upsampling combined with depthwise separable convolutions to decode a 64-dimensional latent vector into a 384×512 grayscale image.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937464API
curl https://kongchang.com/api/v1/knowledge/claims/937464MCP
get_claim(id=937464)