How To Choose Quality Pictures For AI Training

Imagine you want to create a magical machine that can draw pictures of people that look just like the photos you give it. That's essentially what we're doing when we train an AI for portrait generation. But the secret to making this magical machine work well is all in the pictures you choose. Let me walk you through how to pick the perfect pictures.

Think of Training Your AI Like Teaching a Child to Draw

Just like you'd teach a child to draw by showing them many examples, you'll be teaching your AI how to create portraits by showing it carefully selected pictures. The more thoughtful you are about these pictures, the better your AI will learn.

Step 1: Gathering Your Pictures

First, collect high-quality photos of the person you want your AI to be able to draw. Here's what makes a good picture:

• Clear, sharp images (no blurry or fuzzy photos)
• Pictures taken from different angles
• Photos showing the person's full body, head and shoulders, and close-up face shots
• Images where the person looks like themselves (no crazy costumes or extreme makeup)

Tip: Aim for 10-20 really good pictures. Quality matters more than quantity!

Step 2: Choosing the Right Pictures

Imagine you're teaching someone to recognize your friend. You'd want pictures that show:

• What they look like from different sides (front, side, maybe slightly turned)
• Their typical facial expressions
• Their standard clothing style
• Unique features like a distinctive smile or the way they hold themselves

Bad examples to avoid:

• Group photos where a person is just one of many people
• Pictures with busy backgrounds
• Photos with watermarks or logos
• Images that look very different from how the person usually looks

Step 3: Preparing Your Pictures

This is like ironing clothes before putting them away - you want everything neat and organized:

• Make sure all pictures are about the same size (preferably square-shaped)
• Remove any backgrounds that might distract from the person
• Ensure the person is the main focus of each image

Technical Behind The Scenes Made Simple

Our software will do something called "bucketing" - think of it like sorting photos into perfect-sized frames. The goal is to make sure all your images are consistent and easy for the AI to understand.

Training an AI to draw portraits is a bit like teaching someone to paint. It takes time, careful instruction, and lots of good examples.

Recap

To get the best results, remember to:

1. Start with crystal-clear, high-resolution photos
2. Upload both portrait and full body shots of the person
3. Use up to 26 portrait pictures of the person. Preferably cropped to 1:1 aspect ratio (aka square crop).
4. Ideally Use 6 photos of full body or entire object + 10 medium shot photos from the chest up + 10 close-ups.
5. Variation is key - Change body pose for every picture, use pictures from different days backgrounds and lighting. Every picture should introduce new information about the person.
6. Avoid pictures taken at the same hour/day. For example few pictures with the same shirt will make the AI learn the shirt as well as part of the person.
7. Always pick a new background.
8. Do not upload pictures mixed with other people
9. Stay away from pictures that don't feel like the "real" person