Unverified50% confidenceSolutionExact time
基于InsightFace或MediaPipe的面部检测节点先定位人脸边界框，再对每张脸单独执行img2img重绘，最后融合回原图
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRetinaFace是2019年由InsightFace团队提出的单阶段人脸检测器，联合进行人脸检测、关键点定位和3D人脸重建75% similarUnverifiedComfyUI中的Face Detailer通常通过Impact Pack插件实现，采用检测-裁剪-重绘-融合的三阶段流程解决全身镜头中人脸分辨率不足问题74% similarUnverified该工具的底层处理逻辑是逐帧读取视频，对每一帧使用MediaPipe进行人脸检测获取边界框坐标，然后将表情包图片缩放叠加到人脸位置73% similarUnverifiedPopular face feature extraction models include FaceNet, ArcFace, and Dlib's ResNet model73% similarUnverified随着FaceSwap、DeepFaceLab等开源项目的成熟以及基于Stable Diffusion的换脸工具涌现，部分工具仅需一张照片即可生成可信度较高的动态换脸视频72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/916442API
curl https://kongchang.com/api/v1/knowledge/claims/916442MCP
get_claim(id=916442)