<h3 id="software-lllyasvielwebui-forge"><a target="_blank" rel="ugc" href="https://github.com/lllyasviel/stable-diffusion-webui-forge">Software lllyasviel/Webui-forge</a></h3><h3 id="trained-picture-prompt">trained picture prompt</h3><h3 id="3dguofengillustrationtoontangfuchinese-traditional-hanfuqixiong-ruqunphoto-referenced">【<a target="_blank" rel="ugc" href="https://civarchive.com/models/1021411?modelVersionId=1145374">3D</a>】【<a target="_blank" rel="ugc" href="https://civarchive.com/models/109160?modelVersionId=117606">guofeng</a>】【illustration】【<a target="_blank" rel="ugc" href="https://civarchive.com/models/1069710?modelVersionId=1200683">toon</a>】【tangfu,chinese traditional hanfu,qixiong ruqun】【photo-referenced】</h3>Reg picture prompt【1photo】【1woman32】【7788RE】【7788WH】【ALAN】【FENGJING】【HAILUOKE】【LINGXIAOQIAN】【NIGHTYY】【XIAOZHUANGFOX】【travel】----训练图片提示词与正则化提示词<hr /><h3 id="this-lora-model-is-primarily-trained-on-realistic-images.">This LORA model is primarily trained on realistic images.</h3><h3 id="userealistic">use【realistic】</h3>----这是个以真实为主要训练的LORA 【realistic】是它的主要训练风格<h3 id="there-are-only-200+-training-images-and-3100-regularized-imagesrealistic-style-accounts-for-70percent-training-images-foundational-model-is-noobai-eps1.1-.">There are only 200+ training images and 3,100 regularized images,Realistic style accounts for 70% training images, foundational model is NOOBAI-eps1.1 .</h3>----训练图片只有200+张与3100+正则化图片，真实化风格图片占训练图片的70%，基础模型是noobai-eps1.1<h3 id="however-the-output-image-bears-some-resemblance-to-uengine-or-guofeng2-in-sd1.5.-if-want-more-photographing-it-should-be-usephoto-referenced.">However, the output image bears some resemblance to UEngine or Guofeng2 in SD1.5. if want more photographing, it should be use【photo-referenced】.</h3>----但是输出的图片有点像虚幻引擎或者sd1.5的guofeng2 .照片化就加【photo-referenced】<h3 id="in-hires.fix-avoid-using-karras-noise-as-it-may-lead-to-excessive-noise-or-even-overexposure.">In hires.fix, avoid using Karras noise, as it may lead to excessive noise or even overexposure.</h3>----在hires.fix 别用Karras 噪声会多甚至过曝.<h3 id="the-lessstylegreater-of-the-website-linked-below-in-is-flexible.">The &lt;style&gt; of the website linked below in is flexible.</h3>----与下面&lt;STYLE&gt;网址的风格可变,<h3 id="it-doesn't-have-to-be-realistic.">It doesn't have to be 【realistic】.</h3><h3 id=".">是除了真实风格以外其他风格均可适配.</h3>&lt;STYLE&gt;## <a target="_blank" rel="ugc" href="https://300huafeng.vercel.app/single_artist_style.html">https://300huafeng.vercel.app/single_artist_style.html</a>## <a target="_blank" rel="ugc" href="https://300huafeng.vercel.app/mixed_artists_style.html">https://300huafeng.vercel.app/mixed_artists_style.html</a>

Re3mix realistic by Noobai

Deleted some regularization prompts,model_train_type = "sdxl-lora"pretrained_model_name_or_path = noobaiXLNAIXL_epsilonPred11Version.safetensorstrain_data_dir = #######################reg_data_dir = #######################prior_loss_weight = 0.7resolution = "1248,1248"enable_bucket = truemin_bucket_reso = 1024max_bucket_reso = 3840bucket_reso_steps = 64bucket_no_upscale = trueoutput_name = #######################output_dir = #######################save_model_as = "safetensors"save_precision = "bf16"save_every_n_epochs = 1save_state = truemax_train_epochs = 20train_batch_size = 8gradient_checkpointing = truegradient_accumulation_steps = 2network_train_unet_only = falsenetwork_train_text_encoder_only = falselearning_rate = 0.00016unet_lr = 0.00016text_encoder_lr = 0.00016lr_scheduler = "constant"lr_warmup_steps = 0optimizer_type = "AdamW8bit"network_module = "networks.lora"network_dim = 96network_alpha = 48log_with = "tensorboard"logging_dir = "./logs"caption_extension = ".txt"shuffle_caption = falsekeep_tokens = 0max_token_length = 400noise_offset = 0.08color_aug = falseflip_aug = falserandom_crop = falseseed = 1337mixed_precision = "bf16"full_bf16 = truexformers = truelowram = falsecache_latents = truecache_latents_to_disk = truepersistent_data_loader_workers = truevae_batch_size = 8ddp_gradient_as_bucket_view = falseno_metadata = falselr_scheduler_num_cycles = 1network_args = ["rank_dropout=0.25","module_dropout=0.15","conv_dim=64","conv_alpha=32"]USE 30.5G VRAM

Files with SHA256: bf27c9c892c759f0abc8d1dbfd8c12aabe1414d2b91ed31b63972afddcbf63d2