Model training refers to the process of fine-tuning an AI generation model on a specific set of images or videos to teach it a particular subject, style, or visual concept. Rather than relying solely on a model's general knowledge from its original training, custom model training allows creators to teach the model what a specific person, character, product, or aesthetic looks like, enabling consistent generation of that subject across many outputs.

The technical process involves presenting the model with a curated set of reference images or video frames, along with information about what those examples represent. The model adjusts its weights to associate the provided visual information with the concept being taught, learning to reproduce that subject reliably when prompted. Effective training typically requires a varied but focused set of reference images that show the subject from multiple angles, in different lighting conditions, and in varied contexts, giving the model enough information to understand the subject's essential visual characteristics rather than overfitting to a single reference image.

On Morphic, Model Training enables creators to build custom models trained on their own visual references, making it possible to generate consistent characters, branded subjects, or specific stylistic looks across a full project. This is particularly valuable for narrative content requiring the same character across multiple scenes, product marketing where visual consistency with real-world objects is essential, or any creative project where a distinctive look needs to be reliably reproduced throughout.