Doll draft generation method and device based on diffusion model and storage medium

The embodiment of the invention provides a doll draft generation method and device based on a diffusion model and a storage medium. The method comprises the steps that an image containing a human face is acquired; inputting the face image into a feature recognition model for appearance feature recognition to obtain face image recognition features; filling a cue word template with the face image recognition features to generate corresponding forward cue words and reverse cue words; and respectively inputting the forward cue word and the reverse cue word into a pre-trained doll generation model to obtain a doll picture containing the face features, the doll generation model being obtained based on a diffusion model. According to the embodiment of the invention, the method can generate the doll draft image with the face features, improves the similarity between the doll draft image and the face image, and meets the demands of a user. 本公开实施例提出了一种基于扩散模型的娃稿生成方法、装置及存储介质，所述方法包括：获取包含人脸的图像；将人脸图像输入至特征识别模型进行样貌特征识别，得到人脸图像