Unconditional　generation　of　lace　images　using　Generative　Adversarial　Networks　(GANs),　such　as　StyleGAN,　often　results　in　asymmetry　and　messy　textures.　Introducing　conditional　information　into　GANs　can　improve　the　quality　of　generated　images.　However,　existing　conditional　GANs　mainly　focus　on　utilizing　text　conditions　to　enhance　model　performance　and　fail　to　effectively　utilize　image　conditions　to　guide　the　generation　process.　The　goal　of　GANs　inversion　is　to　find　the　exact　latent　code　of　the　given　image　in　the　latent　space　of　GANs.　Hence,　we　incorporate　the　notion　of　inversion　into　the　image　generation　process　of　StyleGAN.　In　this　paper,　we　propose　a　StyleGAN　conditional　lace　image　generation　method　based　on　an　inversion　encoder,　which　adds　a　multi-layer　feature　extraction　mapping　encoder　to　the　original　StyleGAN　model,　and　generates　an　image　with　the　desired　semantic　information　by　inputting　the　lace　image　into　the　encoder　to　guide　the　training　of　the　StyleGAN　generator.　Firstly,　we　use　the　encoder　to　invert　the　lace　image　into　the　StyleGAN　latent　space　and　mix　it　with　random　latent　codes　with　random　probability.　Secondly,　we　improve　the　loss　function　by　introducing　symmetric　loss　and　dynamically　varying　reconstruction　loss.　Finally,　we　demonstrate　the　effectiveness　of　our　approach　by　generating　lace　images　conditioned　on　real　images.　By　inputting　real　lace　images,　we　generate　lace　images　with　controllable　styles.　Our　generated　results　also　outperform　other　generative　adversarial　network　models　in　randomly　generating　lace　images.　©　2024　Copyright　held　by　the　owner/author(s).　Publication　rights　licensed　to　ACM.