03-Preparazione delle Immagini Avatar - SCAI028

In this lesson we will see how to prepare your avatar's images for Lora training First we need to collect a good number of images of our character before starting the training Of course you must first create and develop your avatar following the methods shown in the previous modules How many images are needed Approximately 30-40 high-quality images of your character Of course the more variety the better So I'm talking about close-ups medium shots full body front profile smile neutral expression different poses and angles What do we do if your character is fake Correct observation if your avatar is fictional you certainly can't take a photo shoot of it Then what we will do is go back to the videos and images already created in the previous modules and take screenshots So make sure there are no distortions strange hands or extra limbs in the images which are common artifacts of AI effects Let's look at a practical example Download and open one of your previous videos from clean Choose one where your avatar moves a lot or makes many different expressions Pause at a point with a good angle use a screenshot tool and then save the image You can save them in a local folder or create one on Google Drive Pause it it has a new angle take a screenshot and save it As long as you have a nice variety of images Here is my example I am creating an image folder for Candy's training which is my avatar I have about 46 with all different expressions poses and angles as we can see Petratto full body shoulders close-up sitting walking To help the model recognize and learn follow these guidelines Use the trigger word in the file name Helps the AI to associate the image with your character For example CandyFox-01 jpg Let's see the example right here Use underscores not spaces This helps us avoid technical issues We don't create the file this way where there's a space but we put an underscore Add descriptions Obviously it's optional but can help CandyFox-01-smiling CandyFox-01-glasses CandyFox-01-profileview etc It mainly helps in recognizing the pose or angle Maintain consistency in names so don't change style halfway Confuse the model and use only jpeg or png as the image format I always use PNG so convert other formats before starting What to avoid Let's see Do not mix different characters in the same folder Avoid overly long names and no emojis or special characters Zip the folder Once you have created the folder with all the different angles You need to rename the folder with your trigger word Right click send to then click compressed folder You are now ready for the training in the next module Let's review it Send a compressed folder And here is the trigger word the folder and the zipped file