Step 3: Preparing your Training Image Set

After you have gone through the questionnaire, it is now time to build your training dataset before training.

You'll upload approximately 30-50 images of your character.

These images teach the model the complete range of poses, expressions, environments, and stylistic elements it needs to reproduce your character accurately.

What makes for a good training image?

To ensure the highest quality results, focus on including:

Clear, High-Quality Character Shots

• Face clearly visible

• Minimal motion blur

• Good lighting

Full Body & Action Shots

• Standing, sitting, running, jumping

• Arms and legs visible

• Dynamic poses

Variety of Angles

• Front view

• Side profiles

• Three-quarter angles

• Back view

Range of Expressions

• Happy, sad, surprised, angry, confused

• Neutral expressions

Diverse Environments

This helps the model learn:

• Lighting differences

• Art style

• How the character interacts with space

The model learns best from a wide span of visual information.

Common Mistakes to Avoid

• Uploading images where the face is partially hidden. Makes it hard to identify expressions.

• Too many images of the same pose or angle. Ai’s learning ability is dependent of the type of datasets.

• No props in the environment or no environment present at all, way more chances for AI to hallucinate on the latter. Environment helps Ai understand style and background style that the character lives in.