👌👌👌sUper eAsy Model training 模型训练教程


Updated:

This time I want to train an image of Optimus Prime from the transformers one movie

这次我想训练一个 transformers one电影中擎天柱的形象

The first step is to create a training set of images. You can get a lot of materials by searching on the Pinterest website.

首先是图片训练集的制作,到Pinterest网站去搜索,可以得到大量素材

Go to the tensor website, select online train, and then upload the image

进入tensor网站,选择online train ,然后upload image

Select the image to edit tags and remove the tags that are most relevant to the character in the image. This is counterintuitive, but it must be done

选择图片编辑tag,去掉图片中和角色最相关的tag,这点事反直觉的,但必须如此

Select the model. I chose DreamShaper 2.5D here.

选择模型,我在这里选择的是2.5d的dreamshaper

Write a trigger word. It is best to create a combination of your own, such as "rb123”

写一个Trigger words,这里最好是自己创作的组合,例如“rb123”

Write Model Effect Preview Prompt, which can be Trigger words➕Simple action

写上Model Effect Preview Prompt,可以是Trigger words➕简单动作

Click start train below to start training!

点击下面的 start train就可开始训练啦!

The training progress can be viewed in the training section of the personal center.

训练进度可以在个人中心的training中查看

The default training process is 10 rounds. In each round, there will be 4 pictures previewed. The top has the loss value of each epoch. Generally, the lowest loss value performs best.默认的训练过程是10轮,在每一轮会有4张图片预览,最上方有每个epoch的loss值,一般情况下最低的loss值表现是最好的

选择你认为最好最相似的一个epoch,点击发布,就拥有你自己的lora模型啦!谢谢你的阅读,有不明白的部分,欢迎留言询问

Choose the epoch you think is the best and most similar, click publish, and you will have your own lora model! Thank you for reading. If you have any questions, please leave a message.

2
0