A Seelab model learns your product, your character or your style from a few images. Every visual you then generate with it stays true to the original: same shape, same label, same face, same look and feel. This is what lets you produce hundreds of brand visuals without a photo shoot.

Three types of models

Portraits & Avatars
Recognizable, consistent characters: mascots, virtual models, brand ambassadors.

Products & Objects
Your products, faithful visual after visual, for e-commerce, social media and campaigns.

Styles & Photography
Your brand's visual DNA, in photography and illustration alike: palette, lighting, composition.

What you can do with a Seelab model

• Generate images: choose your model in the model selector and describe the scene.
• Generate videos: one training creates both your image model and your video model.
• Insert your product into an existing image, with Replace in the editor.
• Upscale an image with precision, by selecting your model in Magnify.
• Combine your model with reference images: the product comes from the model, the mood from the reference.

How it works, in short

1. You choose the model type.
2. You import a few images of your product, character or style, and add a brief if needed.
3. Seelab trains the model.
4. The model appears in the model selector of every project member.

Each step in detail: Create and train a model.

Glossary

• Seelab model: a model trained on your products, characters or styles.
• Dataset: a model's training images.
• Training: the step during which Seelab learns your subject from the dataset.
• Project: a team's shared space; its members have access to the same models.

Updated on: 31/10/2026