Loading…
Loading…
Tongyi-MAI/Z-Imagegirlslike_zi_HanGaIn.safetensors 42 MBgirlslike_zi_ShinYuNa.safetensors 42 MBgirlslike_zi_YuJiMin.safetensors 42 MBgirlslike_zi_aimi.safetensors 42 MBgirlslike_zi_ayx.safetensors 42 MBgirlslike_zi_ayxuan.safetensors 42 MBgirlslike_zi_bb.safetensors 42 MBgirlslike_zi_bl.safetensors 42 MBgirlslike_zi_bsn.safetensors 42 MBgirlslike_zi_cdr.safetensors 42 MBgirlslike_zi_ch.safetensors 42 MBgirlslike_zi_chhao.safetensors 42 MBgirlslike_zi_chshu.safetensors 42 MBgirlslike_zi_chy.safetensors 42 MBgirlslike_zi_chyao.safetensors 42 MBgirlslike_zi_cy.safetensors 42 MBgirlslike_zi_cyan.safetensors 42 MBgirlslike_zi_cyx.safetensors 42 MBgirlslike_zi_czh.safetensors 42 MBgirlslike_zi_djie.safetensors 42 MBgirlslike_zi_fbb.safetensors 42 MBgirlslike_zi_fwf.safetensors 42 MBgirlslike_zi_gbt.safetensors 42 MBgirlslike_zi_gfl.safetensors 42 MBgirlslike_zi_gl.safetensors 42 MBgirlslike_zi_glnz.safetensors 42 MBgirlslike_zi_guoxt.safetensors 42 MBgirlslike_zi_gxt.safetensors 42 MBgirlslike_zi_gy.safetensors 42 MBgirlslike_zi_gyj.safetensors 42 MBgirlslike_zi_gyy.safetensors 42 MBgirlslike_zi_gzl.safetensors 42 MBgirlslike_zi_hke.safetensors 42 MBgirlslike_zi_hmt.safetensors 42 MBgirlslike_zi_hmy.safetensors 42 MBgirlslike_zi_hnkz.safetensors 42 MBgirlslike_zi_hq.safetensors 42 MBgirlslike_zi_hry.safetensors 42 MBgirlslike_zi_hsiy.safetensors 42 MBgirlslike_zi_hsy.safetensors 42 MBgirlslike_zi_hx.safetensors 42 MBgirlslike_zi_hy.safetensors 42 MBgirlslike_zi_iu.safetensors 42 MBgirlslike_zi_jc.safetensors 42 MBgirlslike_zi_jjw.safetensors 42 MBgirlslike_zi_jjy.safetensors 42 MBgirlslike_zi_jqq.safetensors 42 MBgirlslike_zi_js.safetensors 42 MBgirlslike_zi_jsy.safetensors 42 MBgirlslike_zi_jt.safetensors 42 MBgirlslike_zi_jwl.safetensors 42 MBgirlslike_zi_jx.safetensors 42 MBgirlslike_zi_jzh.safetensors 42 MBgirlslike_zi_keb.safetensors 42 MBgirlslike_zi_khw.safetensors 42 MBgirlslike_zi_lc.safetensors 42 MBgirlslike_zi_lhc.safetensors 42 MBgirlslike_zi_ljx.safetensors 42 MBgirlslike_zi_ljy.safetensors 42 MBgirlslike_zi_lld.safetensors 42 MBgirlslike_zi_llz.safetensors 42 MBgirlslike_zi_lq.safetensors 42 MBgirlslike_zi_lqx.safetensors 42 MBgirlslike_zi_lrt.safetensors 42 MBgirlslike_zi_lsf.safetensors 42 MBgirlslike_zi_lt.safetensors 42 MBgirlslike_zi_lwd.safetensors 42 MBgirlslike_zi_lxp.safetensors 42 MBgirlslike_zi_lxran.safetensors 42 MBgirlslike_zi_ly.safetensors 42 MBgirlslike_zi_lyan.safetensors 42 MBgirlslike_zi_lyer.safetensors 42 MBgirlslike_zi_lyf.safetensors 42 MBgirlslike_zi_lyt.safetensors 42 MBgirlslike_zi_lz.safetensors 42 MBgirlslike_zi_lzl.safetensors 42 MBgirlslike_zi_msc.safetensors 42 MBgirlslike_zi_mxt.safetensors 42 MBgirlslike_zi_myl.safetensors 42 MBgirlslike_zi_mzy.safetensors 42 MBgirlslike_zi_nn.safetensors 42 MBgirlslike_zi_ql.safetensors 42 MBgirlslike_zi_qsz.safetensors 42 MBgirlslike_zi_qw.safetensors 42 MBgirlslike_zi_saj.safetensors 42 MBgirlslike_zi_sc.safetensors 42 MBgirlslike_zi_sff.safetensors 42 MBgirlslike_zi_sl.safetensors 42 MBgirlslike_zi_sml.safetensors 42 MBgirlslike_zi_suss.safetensors 42 MBgirlslike_zi_sy.safetensors 42 MBgirlslike_zi_syi.safetensors 42 MBgirlslike_zi_syn.safetensors 42 MBgirlslike_zi_syq.safetensors 42 MBgirlslike_zi_szer.safetensors 42 MBgirlslike_zi_tly.safetensors 42 MBgirlslike_zi_txw.safetensors 42 MBgirlslike_zi_ty.safetensors 42 MBgirlslike_zi_wcr.safetensors 42 MBgirlslike_zi_wjn.safetensors 42 MBgirlslike_zi_wq.safetensors 42 MBgirlslike_zi_wqw.safetensors 42 MBgirlslike_zi_wxc.safetensors 42 MBgirlslike_zi_wy.safetensors 42 MBgirlslike_zi_wzx.safetensors 42 MBgirlslike_zi_xjq.safetensors 42 MBgirlslike_zi_xly.safetensors 42 MBgirlslike_zi_xq.safetensors 42 MBgirlslike_zi_xrh.safetensors 42 MBgirlslike_zi_xrx.safetensors 42 MBgirlslike_zi_xxy.safetensors 42 MBgirlslike_zi_ycy.safetensors 42 MBgirlslike_zi_ygr.safetensors 42 MBgirlslike_zi_yjy.safetensors 42 MBgirlslike_zi_ymn.safetensors 42 MBgirlslike_zi_yr.safetensors 42 MBgirlslike_zi_yrong.safetensors 42 MBgirlslike_zi_ysx.safetensors 42 MBgirlslike_zi_yy.safetensors 42 MBgirlslike_zi_yyongyi.safetensors 42 MBgirlslike_zi_yzm.safetensors 42 MBgirlslike_zi_zbz.safetensors 42 MBgirlslike_zi_zch.safetensors 42 MBgirlslike_zi_zhm.safetensors 42 MBgirlslike_zi_zhmei.safetensors 42 MBgirlslike_zi_zjn.safetensors 42 MBgirlslike_zi_zjning.safetensors 42 MBgirlslike_zi_zjy.safetensors 42 MBgirlslike_zi_zl.safetensors 42 MBgirlslike_zi_zls.safetensors 42 MBgirlslike_zi_zlt.safetensors 42 MBgirlslike_zi_zm.safetensors 42 MBgirlslike_zi_zmy.safetensors 42 MBgirlslike_zi_zs.safetensors 42 MBgirlslike_zi_zti.safetensors 42 MBgirlslike_zi_zxd.safetensors 42 MBgirlslike_zi_zxt.safetensors 42 MBgirlslike_zi_zxy.safetensors 42 MBgirlslike_zi_zy.safetensors 42 MBgirlslike_zi_zye.safetensors 42 MBgirlslike_zi_zyx.safetensors 42 MBgirlslike_zi_zyz.safetensors 42 MBgirlslike_zi_zzx.safetensors 42 MBgirlslike_zi_zzy.safetensors 42 MBgirlslike_zi_cdl.safetensors 42 MBgirlslike_zi_chg.safetensors 42 MBgirlslike_zi_jpy.safetensors 42 MBgirlslike_zi_lss.safetensors 42 MBgirlslike_zi_szn.safetensors 42 MBgirlslike_zi_wys.safetensors 42 MBgirlslikezit_cdl.safetensors 35 MBgirlslikezit_gxt.safetensors 35 MBgirlslikezit_hnkz.safetensors 35 MBgirlslikezit_jt.safetensors 35 MBgirlslikezit_lyf.safetensors 35 MBgirlslikezit_syn.safetensors 35 MBgirlslikezit_szer.safetensors 35 MBgirlslikezit_tly.safetensors 35 MBgirlslikezit_ty.safetensors 35 MBgirlslikezit_wcr.safetensors 35 MBgirlslikezit_ycy.safetensors 35 MBgirlslikezit_zjn.safetensors 35 MBgirlslikezit_zjy.safetensors 35 MBgirlslikezit_bl.safetensors 35 MB用于测试基于 Z-Image Base/Turbo BF16 训练的人物 LoKR / LoRA，仅代表个人审美与训练效果。实际生成结果仅供测试，请勿用于冒充、欺骗或其他不当用途。
后缀为zi代表基于z-image base训练，zit则代表基于turbo训练，zi在base与turbo模型中均可使用，zit则只能在turbo中使用。
z-image base训练尚无较为通用的可行方案，目前本zi系列选择的方案是全秩lokr，可在与turbo模型配合使用时以权重1.0出图，也能适当增加权重以增强相似性，最好不要超过1.5。
而在与base模型或相关微调模型配合使用时，可适当降低权重。
基于base训练的zi lokr在turbo模型中使用时，质感会更接近turbo本身（应该是base模型本身低噪部分难以被"污染"的缘故），中远景图中的人物相似度也会比zit好一些，近景则更"平淡"。
但总体来说兼容性存疑，比如解剖学问题仍然大量存在（在base模型上生图就存在类似问题，也影响到了训练，会让肢体问题重新变得不稳定又棘手），其他类似的问题都还在测试中，后续有更好的方案会继续更新。
示例图生图参数设定：
z-image turbo bf16；
8步；
er_sde + sgm_uniform | euler + flowmatch；
cfg=1；
lora权重0.5~1.5；
分辨率1080*1440；
中文提示词；
无其他lora参与。
鉴于z-image turbo模型本身就能通过人名提示词直出某些真实人物，因此本系列lora训练中也添加了人物本名作为触发词，生图时需输入触发词才能准确画出对应人物。（名字本身就是触发词，无需其他前缀后缀；有些人物在base模型中是无法直出的，但在turbo模型中是可以的，要注意甄别）
如果原模型触发词已被占用，则使用其他特定触发词，特定触发词会包含在示例图的水印中（水印通常以girlslikezi for xxx形式出现，xxx即为触发词），有水印的示例图就代表需要使用特定触发词而非本名。
（许多真实人物已经在z-image的模型数据集中了，只是权重没有个别人物那么高，为了避免人物概念的混淆，也只好添加触发词来规避或加强生成）
z-image turbo的一些注意事项：
4步就初步可用，10步以上没有明显收益
多样性生图可使用双采样器，第一步采样时调度器改为ddim可以得到不错的效果
同样可以使用ModelSamplingAuraFlow节点设定shift值来更改出图效果，3-10之间均可，shift=3时与原图一致
0.7<cfg<1时会放大z-image本身的质感优势，3>cfg>1时会更准确但画风会更“AI"，两者均会增加生成时间
基于turbo训练的lora多兼容性不太好，用base训练的lora多兼容性会好很多
示例图片一如既往包含工作流，zi示例图lora强度为1.0，zit示例图lora强度为0.9，均由z-image turbo bf16模型生成。
示例图人名缩写可在 girlslike_lora_gallery_app 中查阅对应真名，此缩写girlslike系列通用。
本仓库展示的绝大部分例图提示词，均引用自 Civitai 平台的 【中文】【SFW】15000+写真 & 自拍通配符，特此鸣谢。