Have you ever tried using AI to generate your own look? Have you ever felt others' are in good quality, but not yours? Have you ever tried to follow tutorials to generate but it’s too difficult and complicated you can't handle it at all? More powerful than Photoshop, something out of nothing AI generated photo tutorial, now begin! （Sorry, I'm sick again. My voice is a bit hoarse） hello everyone! Welcome to "An IT-a", I'm Anita Anyone who follows me on IG should know this I always like to play AI tools with my own look For example, photo generation, video to animation, AI spokesperson, DeepFake... The entry-level ones should be photo generation and DeepFake There are many websites and apps on the market that can do this However, you can only generate styles prepared by the them. No freedom in creation, not able to retry Having seen this I specially prepared a method You don’t need to have much IT knowledge but can generate this bunch of AI photos You can also change the prompt to suit your own preferences. to generate your own beautiful AI photos My method is completely free And you don’t need to install anything on your computer Very suitable for beginners The only thing needed is a free Google Account If you use Gmail Then you can start playing I have also prepared a bunch of sample prompts for everyone to start with. What are you waiting for? Let's start immediately First, let me talk about the method I will use this time. My method is to use more than ten of your photos for training Then generate a new AI photo You may ask Why sometimes a website or app can do this with just one photo? In fact, their method is called DeepFake The whole approach is different from that of Training DeepFake is taking a photo of you Merge your face into another photo So you can’t use Prompt to control photos And it will only replace your eyes, nose, and mouth So if the contour of the faces vary too much You will see that it is very hard to put your face in The effect will not be good But the advantage is that it needs fewer photos one is enough And if you have a favorite photo DeepFake is really fast and pretty The first thing to do in training This is also the only difficult part of the whole process. Is to prepare your photos I strongly recommend keeping the quantity at 15 But if you really don’t have as many as 15 There must be at least 10 Remember! What you are doing in this step is to teach the AI how you look like Remember this and you will understand how to choose the right photo First, I'll recommend you to give all headshots Unless your body shape is very special you really want it to learn it Otherwise, I would recommend giving all as headshots If you really want to give full body shots just give one or two otherwise your face will not be alike And if you really want to give full body shots your clothes should be tight-fitting If you wear loose clothes your body shape will be covered up the AI won't know how your body shape looks like it’s useless just misled him Second Unless you want it to learn what you look like after disguising otherwise Just don't give it some photos of you in disguise. Remember you want it to know what you look like So your appearance needs to be clear Don't have anything blocking your eyes, ears, mouth and nose Third Don’t make your photos underexposed or underexposed It's like going to Lan Kwai Fong You must go to a certain tea restaurant for late night snacks not because you're hungry It's to see the face clearly Fourth Do not include other people in your photos otherwise, it won't know who is you Fifth Include as many different angles as possible in your photos It's all that sentence In order to let him know clearly what you look like Here is the Data Set I gave I have two in my pajamas No makeup, casual photo shoot No one will see the photos you use So don’t worry about the photos not being pretty We will beautify it during the generation step. Don't worry After selecting the photo You can start doing Resize Resize to 512x512 size We usually use the cropping tool - Birme Drag your 15 photos in Then it will automatically help you find your face but sometimes I find the wrong one Then do it yourself Then check that the height and width are both 512 Then you can press Save Choose Zip or Files It doesn't matter If you Zipped it, Unzip it Open the folder Then we have to describe your photo to the AI First we need to create a txt file If you are on a Windows machine Just use NotePad Remember to save as Plain Text It’s a bit more troublesome if you’re using a Mac First open a TextEdit Click Settings in the menu in Format, select Plain Text Then the file you create will become a txt file Let’s take this photo as an example Create a txt file with the same name First we have to give the character a name Use a word that doesn't exist The easiest way is to use your username I will use "An IT-a" Remember what you used here We will reuse it later Give a comma after Describe your photo The more detailed the better For example, this is what I have Separate each description with a comma If your photo has other things in it Just describe it For example, this The principle is Describe your hairstyle, clothing, action, objects and backgrounds Do this with every photo Then congratulations to you The hardest part is done Then it can be handed over to AI Open the first link in my description column You will see a colab First we press "save a copy in drive" in "file" You will clone a copy to your own google drive There is a downward arrow in the upper right corner Click "Change runtime type" Choose "T4 GPU" Then press "save" Press the "run" below step 1 You will be asked for permission to connect to your google drive grant it If you are using it for the first time It may ask you a few more things Anyway, if you have something to tick, just tick it. If you have something to allow, allow it Then step 2 will open a file called "MyLoras" folder to do this If you want to use another folder name Just change it here second line Change to the username you just used for the txt file Then press run give it some time After finished, run step 3 and step 4 Then go to google drive Open the created folder There is a "train_data" folder inside, open it Then upload the pile of pictures and txt file you prepared Wait until the upload is complete Run step 6 Finally run step 7 This step is the real training Will wait a little longer Usually about half an hour after completion You will see that there are 5 new files created in the output folder. Now you can start generating photos Open the second link in the description column It will be a colab that generates photos Same as before Press "save a copy in drive" in "file" Press the down arrow in the upper right corner change "change runtime type" to "T4 GPU" press save Run step 1 Grant permission to use Google Drive Run step 2 and step 3 Then fill in the username you used before version, usually the best is 0003 What we are talking about here are the 5 files created you can also try other versions See which one works best drive path Change back your google drive path Remember to change your username Then we can choose our model I've prepared 3 in advance realistic, 2.5D and 3D three styles You can also go to this website Find the model you want to use Just copy the download link paste it and it can be used Then choose what style your model is Basically, it’s the difference between non-anime and anime Then if you use a model of your own The model description wrote that it's baked VAE Then tick this Finally, there are prompt and negative prompt I have prepared 4 templates per model here for your reference You can change it to match your needs I guess everyone knows English It should be easy For example, you can add an action, holding something, less details, changing colors... If you study more You can tune the following parameters If not, I would suggest not touching it. press run If you are using the model for the first time, you'll need to wait a little longer for it to download If you are not satisfied with the result You can run this step again Run until satisfied For colab’s free plan although there are limitations But it's actually hard to reach it's limit Feel free to use it If it really says that you have used up your quota Wait a while or a day It can be used again If you are satisfied, run the next step to enlarge it Just click right on it save it Then you're done If you want to be generate again next time Keep the newly created google drive folder then start with the inference colab directly but if the connection is lost Then you have to start over from the connect step. Anyone who plays stable diffusion is used to it. Is that the hands and fingers are usually failure Sometimes if you generate a photo you're happy with but the fingers failed There is a way Enlarged it I apologise here AdobeExpress was free before but it start to require subscription recently So I found a replacement instead This is the inpaint tool that I have found with the most free credits if you have a better inpaint tool tell me in the comments below Let’s go to the image editor in getimg.ai Paint your fingers Write "hands" in prompt press "inpaint" If you are not satisfied, you can press cancel and try again There are 100 credits per month each trial uses 4 credits That means you have 25 free opportunities every month If you tried many times, but still can't get a good one there is another way cover it up! For example, change your prompt to "a bunch of flowers" Done Remember! It's normal for Stable diffusion to take many attempts to get a satisfactory photo Don't stop just because you're not getting good results for the first time try more times You will see beautiful photos Remember What I am using here is version 1.5 of stable diffusion for training So we need to use the model of version 1.5 When you look for your model, you can filter it like this I myself have tried doing the whole thing with sdxl But the effect is not satisfactory yet So I’ll keep using 1.5 for the time being. In addition, my demonstration today was face of girls if you want me to make templates for boys Leave a message below to let me know I wrote so much code And shared so many templates You would give me a like, right? If you haven't subscribed to my channel Subscribe now In addition, if you like this type of AI tutorial but haven't watched my my AI customer service chatbot tutorial click here now See you next time! BYEBYE!
