LoRa Lab _ see-through _ pantyhose _ 3 pic _ without reg

Main Lora of this series is ...👉👉👉 https://civitai.com/models/1360230?modelVersionId=1536616Intro.This series of LoRa is used to compare the effects and differences of LoRa trained under different schemes.Don't have too high expectations for them, but on the contrary, if you're interested, you can download the attached dataset and try it yourself. With the parameters below, except for the 24 images training, the training is rather easy and quick.NOTE Future content update plans:Upload LoRa results for stages 02, 04, 06, 08 (final), note that wr means With Reg, wor means Without RegUpload dataset (image & caption) & traing logs (tensorboard)Update links to other management experiment models on civitUpload some test imagesLater, one of the LoRa will be selected as the "main LoRa", and some subsequent comparisons and ideas will be updated thereExplanation of LoRa in Different SchemesTry the following schemes to see how the trained "nsfw stylized (pantyhose & see-through)" LoRa feels and what the differences are:1 image, no regularization1 image, with regularization3 images, no regularization3 images, with regularization8 images, no regularization8 images, with regularization24 images, no regularization24 images, with regularizationThis time captions are directly derived using WD14 Conv v2 tagger, without any additions or deletionsDesired Training EffectsIncrease the maturity of girlsAdd natural transparency effect to clothingAdjust the style of pantyhose (flesh-colored shiny style; and increase the default probability of pantyhose when not specified)Potential IssuesThe entire dataset is 1girl, solo, so the default style for other characters may have problems (such as male characters)Not sure about LoRa's ability to handle complex promptsCurrent Training SettingsNumber of images: 3 Regularization: No Model Training Parameters[Training Parameters] BASE=https://civitai.com/models/827184/wai-nsfw-illustrious-sdxl EPOCHS=8 # Number of training epochs SEED=1861

[Resolution Settings] TRAINING_WIDTH=1024 # Training width TRAINING_HEIGHT=1152 # Training height MAX_BUCKET_RESO=$TRAINING_HEIGHT # Maximum bucket resolution

[Hardware Adaptation] DIM=32 # Network dimension (3090 configuration) DIM_ALPHA=28 # Network alpha value (3090 configuration) TRAINING_BATCH_SIZE=1 # Batch size

[Training Parameters] --train_batch_size 1 --learning_rate 0.0001 --unet_lr 0.0001 --text_encoder_lr 0.00001 --optimizer_type "AdamW8bit" --mixed_precision "bf16" --save_model_as "safetensors" --save_precision "fp16" --save_every_n_epochs 2 --cache_latents_to_disk --xformers --lowram 介绍本系列的 LoRa 是用于对比不同方案下训练出来的 LoRa 的效果和差异不要对它们有太高预期, 但相反的, 如果你感兴趣, 可以下载附录的数据集自己尝试, 按照下面的参数, 除了 24 张的训练, 其他一次训练时间和成本都非常低NOTE 后面会内容的更新计划 :上传 02, 04, 06, 08 (final) 阶段的 LoRa 结果, 小提示 wr 指的是 With Reg, wor 指的是 Without Reg上传 数据集 (image & caption) & logs (tensorboard)更新与其他管理实验 model 的 civit 的链接上传一些测试图片后续会选定其中一个 LoRa 作为 "主要 LoRa", 一些后续的对比和想法, 会更新到那里去不同方案的 LoRa 说明尝试一下如下的方案, 所训练出来的 "nsfw 风格化 (pantyhose & see-through)" 的 LoRa 感觉如何, 各自有何差别 :1 张图片, 无正则化1 张图片, 有正则化3 张图片, 无正则化3 张图片, 有正则化8 张图片, 无正则化8 张图片, 有正则化24 张图片, 无正则化24 张图片, 有正则化本次 captions 尝试直接用 WD14 Conv v2 tagger 推导, 无任何的增删训练想要的效果提升女孩的成熟度加入衣物的原生透明效果调整连裤袜的风格 (肉色舍宾风格; 且增加无指定时裤袜的默认几率)可能的问题全数据集都是 1girl, solo, 所以其他形象的默认风格可能会出现问题 (如男人形象)不太确定 LoRa 在处理复杂 prompt 的能力本次训练相关设置图片数量: 3 是否正则化: 否 模型训练参数[训练参数] BASE=https://civitai.com/models/827184/wai-nsfw-illustrious-sdxl EPOCHS=8 # 训练轮数 SEED=1861

[分辨率设置] TRAINING_WIDTH=1024 # 训练宽度 TRAINING_HEIGHT=1152 # 训练高度 MAX_BUCKET_RESO=$TRAINING_HEIGHT # 最大分桶分辨率

[硬件适配] DIM=32 # 网络维度 (3090配置) DIM_ALPHA=28 # 网络alpha值 (3090配置) TRAINING_BATCH_SIZE=1 # 批次大小

[训练参数] --train_batch_size 1 --learning_rate 0.0001 --unet_lr 0.0001 --text_encoder_lr 0.00001 --optimizer_type "AdamW8bit" --mixed_precision "bf16" --save_model_as "safetensors" --save_precision "fp16" --save_every_n_epochs 2 --cache_latents_to_disk --xformers --lowram