Step 5: Model Versions & Selection

As your dataset is processed, the system trains multiple versions of your model. Each version represents a slightly different interpretation of your character based on how the training data was learned. This gives you visibility into what the AI understands and control over which version performs best for your IP.

After training, you’ll see several generated versions to review and rate. These versions may differ subtly in things like proportions, expressions, pose accuracy, or how the character interacts with environments. By reviewing and rating them, you’re helping the system learn what worked well and what didn’t.

Once you’ve identified the version that best represents your character, you can select “Use This Version.” From that point on, all image and video generation will be powered by that selected model version, ensuring consistency across your outputs.

Gemini with Contextual Training

One of the available options is Gemini with Contextual Training, which is typically the most accurate. This version places greater emphasis on both character details and environmental context, helping the model better understand how your character should look and behave across different scenes. Because it accounts for more of the surrounding context, it often produces the most stable and on-model results.

Choosing the right version is an important step in finalizing your character. It ensures that the model you use for generation reflects your creative intent and continues to improve as your dataset evolves.