Agentic Creative Flow

The Agentic Creative Flow allows you to train custom AI models on your product images. Once trained, your model generates brand-consistent creative assets based on your text prompts. This capability enables unlimited creative exploration and natural scene integration.

Guides

• Follow our Dataset Preparation Guide to collect a set of images to achieve the best results.
• Follow our Prompting Guide to generate your first creatives.

How It Works

• Select a brand to organize your creative assets. This creates a dedicated workspace where all your models, products, and generated creatives will be stored. Your team members can collaborate within this brand workspace.

• Or, create a new brand if you haven't already.

• Upload your product assets to create a training dataset. Group similar products into categories for better results. Follow our Dataset Preparation Guide to collect a set of assets to achieve the best results.

• Train a custom model specifically for your brand. This process typically takes 8-18 hours depending on dataset size and complexity. Each model is optimized to understand your specific products and brand aesthetic.

• Approve Themes generated by our agentic system. Once training completes, the system automatically suggests creative themes based on your products. Our AI agents analyze your product images to understand key visual characteristics, usage contexts, and potential marketing angles. This intelligence allows the system to suggest relevant, on-brand creative directions without manual brainstorming. For example, if you've uploaded running shoes, the system might suggest themes like "urban fitness," "trail running," or "athlete lifestyle" - each representing different ways to showcase your product that resonate with specific customer segments.

• Generate unlimited creatives using text prompts that incorporate your brand elements. Follow our Prompting Guide to achieve the best results.

• View results from the theme generation. Images are produced in batches for efficient review. Use the like, dislike, and delete functions to organize results. Liked images move to the front of your collection while disliked ones are moved to the end.

• Edit and refine your favorite outputs with natural language instructions. Select any generated image and click the edit button to make specific adjustments. Simply describe the changes you want (e.g. "make the background brighter" or "change the scene to a beach") and the model will create a refined version. You can also adjust the dimensions of your creative to fit specific media placements.

• View edited result and keep iterating.

Limitations

• Products with intricate details or text may show reduced fidelity in generated images
• Training requires 8-18 hours before generation can begin
• Best results require high-quality, consistent product photography

Examples of AI Generations