Zunyi-Business-LORAs/Luxury-liquor-bottle

Luxury-liquor-bottle

<Gallery />

Model description

We used PS to process some images and resized 10 training images to 512*512. We need to remove some brands and logos; and reserve all bottle styles for training. We define this style as luxury and classic white-liquor-bottle. After our training, we want to get this Lora for future research and design.

We updated our prompts for liquor bottle. For now, we could extend our length of prompts without UTF-8 errors. Please check our new prompts and you may find new way to write prompts.

Please use same prompts when you are using this Lora on workflow, which is based on our test results. Example(1. Luxury-liquor-bottle | minimalist shape | Elegant &amp; artistic 2.Luxury-liquor-bottle | Transparent glass | Deep blue-clear gradient )

Many people may need to know trainging speed. In advanced options, we change --saveeverynepochs to 1 (default is 4) . We opened(ticked) --fp8baseunet and --consolelog_simple. And other values are default. We get 1600 default steps. For ten training images, we spent one hour to get Lora and closed all important background processes.

We used doubao to get prompts and input all prompts without Florence-2.

You could check images generated by self-trained Lora on uploaded files.

我们使用了PS把图像处理了一下，把商标，Logo去除来提高训练并保留瓶身的风格，同时在训练以前需要把尺寸固定在512*512。现在我们训练的10张图片定义为高端一些的白酒瓶。通过训练得到一个Lora来为以后的研究和设计做铺垫。我们打算把白酒瓶型进行分类训练，得到不同类型的Lora。

我们同时更新了新的提示词，现在我们的提示词已经延长很多，但是训练时间没有增加，请大家检查新的提示词风格。找找灵感。10张图片的训练时间为一个小时，同时关闭了所有重要进程，

在ComfyUI测试环节的提示词案列（第一：Luxury-liquor-bottle | minimalist shape | Elegant &amp; artistic 第二：Luxury-liquor-bottle | Transparent glass | Deep blue-clear gradient）

打开了高级选项中的 --fp8baseunet 和--consolelogsimple，把saveeveryn_epochs 改为 1，其他全部不变。

我们使用了豆包来获取提示词，手动输入到Fluxgym里，没有用到Florence-2。

注：工作流和自训练的Lora生成样图在附件里。

Trigger words

You should use Luxury-liquor-bottle to trigger the image generation.

Download model

Download them in the Files & versions tab.