Verified65% confidenceReleaseExact time
ControlNet was proposed by Zhang Lvmin in 2023; by attaching additional conditional control branches to a pre-trained diffusion model, it supports multiple types of structured inputs—such as edge maps, depth maps, human skeletal poses, and semantic segmentation maps—to control the composition and spatial layout of generated images.
3
Sources
65%
Confidence
Long-term
Relevance
9/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedControlNet was proposed by Zhang Lvmin et al. in 2023, injecting conditional signals through an additional trainable bypass network without modifying the original model weights.77% similarUnverifiedControlNet通过引入与主干U-Net并联的可训练副本，将姿态骨架图、深度图等结构化控制信号注入生成过程，但对面部特征的精确复现能力有限75% similarUnverified原生多模态预训练在统一token空间中对图像patch和文本token进行联合自回归训练，使模型在权重层面形成跨模态语义绑定68% similarUnverifiedThe pretraining and fine-tuning paradigm originated from NLP and computer vision, with its core being large-scale data pretraining to learn general low-level features, and downstream tasks adapted using a small amount of labeled data.66% similarUnverifiedMultimodal pre-trained models represented by the LayoutLM series jointly model visual, textual, and positional information, advancing the capability frontier in the field of document understanding66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/870462API
curl https://kongchang.com/api/v1/knowledge/claims/870462MCP
get_claim(id=870462)