Create personalized products and marketing campaigns using Amazon Nova in Amazon Bedrock

TL;DR

The Fragrance Lab demonstrates an end-to-end, personalized product and marketing-asset pipeline built with Amazon Nova in Amazon Bedrock.
The Cannes Lions 2025 activation, created with Wildlife, showcased hyper-personalized fragrance development and ad campaign generation, earning Gold and Silver Stevie Awards.
The stack includes Nova Sonic for natural-language interactions, Nova Pro for fragrance design with Retrieval Augmented Generation (RAG), Nova Canvas for visuals, Nova Reel for video, and Polly for voice; Guardrails enforce safety policies.
The experience accelerates perfumers from hours to minutes and enables on-site generation of hundreds of unique fragrances per day, plus customized marketing assets.
While demonstrated for fragrance, the architecture is adaptable to many retail and consumer goods categories, from fashion to food and beverage.

Context and background

The Fragrance Lab is an immersive demonstration that highlights how generative AI can empower hyper-personalized consumer goods and accelerate advertising concept and campaign asset development. It was showcased at the Cannes Lions International Festival of Creativity 2025 as part of AWS’ ongoing exploration of generative AI in retail, consumer products, and marketing. Following the event, The Fragrance Lab received both Gold and Silver Stevie Awards in the Brand & Experiences category, underscoring the impact of the approach. The project was built using Amazon Nova in Amazon Bedrock, illustrating an end-to-end architecture that can be adapted across categories such as fashion, skincare, cosmetics, food and beverage, and home goods. The collaboration with Wildlife was central to translating AWS generative AI capabilities into a tangible physical experience that attendees could interact with. AWS blog. The experience centers on a natural-language interface called Amazon Nova Sonic, which engages attendees in dialogues to understand personality and preferences. Nova Sonic’s conversation state is managed through integrated tools like addTraitTool, removeTraitTool, and uiActionIntentTool to maintain a consistent flow. Collected conversations and trait data feed a custom Retrieval Augmented Generation (RAG) system built with Amazon Nova Pro, a multimodal model that analyzes interactions to extract keywords and determine fragrance notes and composition. Guardrails in Amazon Bedrock provide customizable safeguards to block undesirable topics and ensure a smooth customer experience. Nova Pro translates user inputs into a fragrance recipe, mapping traits to base, heart, and top notes. On-site perfumers then craft hundreds of unique fragrances per day, moving from hours to minutes in the design process. Once a fragrance is defined, Nova Canvas generates customized marketing creative — including fragrance names, taglines, and imagery — which attendees can further tailor. Nova Reel transforms the fragrance image into dynamic video content, and a French-accented voice via Amazon Polly completes the campaign experience. The entire workflow is orchestrated within Amazon Bedrock, illustrating how multiple Nova models can be combined to deliver a cohesive, personalized customer journey. AWS blog. The Fragrance Lab underscores how conversational AI, AI-powered product development, and AI-driven creative generation can be integrated to deliver customized customer experiences. The same underlying architecture can be extended with translations, sizing, and other regional variations as requirements evolve. A video walkthrough of The Fragrance Lab at Cannes Lions 2025 is available for interested readers through the event coverage. The use case is positioned as a blueprint for applying AI across retail and consumer goods categories beyond fragrance.

What’s new

The post highlights the introduction of a fully integrated, end-to-end pipeline built with Amazon Nova in Amazon Bedrock to support hyper-personalized product design and campaign asset generation. Notable updates include:

A modular flow where Nova Sonic handles natural-language interaction and trait/state management via tool integrations (addTraitTool, removeTraitTool, uiActionIntentTool).
Nova Pro serving as the intelligence engine, enabling a Retrieval Augmented Generation (RAG) system that sources real-time fragrance knowledge to inform notes and composition.
A data-flow architecture designed to map user traits to fragrance notes (base, heart, top) and to automate the generation of marketing assets via Nova Canvas and Nova Reel.
Guardrails in Bedrock to block allergens or harmful content, ensuring a safe and compliant customer experience.
On-site perfumers working in tandem with AI to accelerate fragrance personalization and production, delivering hundreds of unique fragrances per day.
Campaign creative that is generated and then customizable by attendees, with options for voice, imagery, and video via Nova Reel and Polly. The approach demonstrates how an end-to-end AI-driven experience can combine natural-language interfaces, AI-generated product concepts, and AI-powered marketing creative to deliver personalized experiences at scale. The architecture is designed to be adaptable across multiple retail and consumer goods categories, including skincare, fashion, food and beverage, home goods, and wellness. AWS blog.

Why it matters (impact for developers/enterprises)

For developers and enterprises, The Fragrance Lab showcases a practical blueprint for achieving hyper-personalization and faster time-to-market for both products and marketing. Key takeaways include:

Hyper-personalization at scale: By combining conversational AI, RAG-powered knowledge, and AI-assisted fragrance design, brands can deliver customized products and campaigns rapidly, moving from concept to customer-facing assets in minutes rather than hours.
Integrated end-to-end workflows: The architecture demonstrates how dialogue management, trait tracking, ingredient mapping, and creative generation can be orchestrated to produce cohesive experiences that span product development and advertising.
Safety and governance: Bedrock Guardrails offer customizable safeguards to help block undesirable topics, supporting responsible AI and regulatory compliance in consumer-facing applications.
Cross-category applicability: While demonstrated in fragrance, the approach can be adapted across fashion, skincare, cosmetics, food and beverage, home goods, and wellness, enabling consistent, AI-assisted customer journeys across categories.
Speed and creativity: On-site AI-assisted perfuming and automated campaign asset generation empower both customers and experts, enabling iterative exploration of ideas and rapid prototyping of concepts.

Technical details or Implementation

The Fragrance Lab’s implementation combines several Amazon Nova capabilities within Amazon Bedrock to deliver a cohesive experience:

Amazon Nova Sonic: A speech-to-speech model that engages with attendees to understand personality and preferences. It maintains conversation state through integrated tools and uses a well-defined prompt to steer interactions.
Tool integrations: addTraitTool, removeTraitTool, and uiActionIntentTool manage user traits, interface actions, and conversational flow to provide a reactive and consistent user experience.
Amazon Nova Pro with RAG: Nova Pro analyzes interactions and extracts keywords to determine fragrance notes. A Retrieval Augmented Generation (RAG) system extends Nova Pro with access to up-to-date fragrance knowledge, ingredient profiles, and strength relationships among ingredients and user identities.
Knowledge sources and fragrance design principles: Real-time access to ingredient knowledge and fragrance design principles informs base, heart, and top-note selections that align with user traits.
Amazon Bedrock Guardrails: Customizable safety and responsible AI policies block allergens and other undesirable content to ensure a seamless customer experience.
Amazon Nova Canvas: Visualizes the resulting fragrance concepts and generates original compositions that reveal fragrance names, ingredients, and a visual identity for marketing assets.
Amazon Nova Reel: Produces dynamic video content from the fragrance visuals, enabling attendees to customize campaigns and download assets.
Amazon Polly: Provides a French-accented female voice for the campaign videos to match Cannes Lions’ atmosphere.
On-site perfumers: AI-generated fragrance formulations are accelerated and then refined by perfumers who finalize hundreds of unique fragrances per day.
Data flow and orchestration: The experience centers on Nova Sonic for dialogue with trait-driven workflows, with Nova Pro performing the fragrance design and Nova Canvas/Reel handling the marketing creative pipeline. The architecture supports translation and sizing enhancements as needed. The flow illustrates how multiple Nova models can be connected to deliver a cohesive, personalized customer journey, from initial conversation to fragrance development and final campaign assets. For reference, see the original AWS post describing The Fragrance Lab and its components: AWS blog.

Key components and roles (at a glance)

| Component | Role in the workflow | Example function / outcome |---|---|---| | Amazon Nova Sonic | Conversational interface | Understands attendee personality and preferences; maintains state via tools |addTraitTool / removeTraitTool / uiActionIntentTool | State and flow management | Manage user traits, trigger appropriate workflows, ensure consistent conversation |Amazon Nova Pro | Intelligence engine with RAG | Analyzes interactions, extracts keywords, determines fragrance notes; RAG sources real-time fragrance knowledge |Retrieval Augmented Generation (RAG) | Knowledge expansion | Extends Nova Pro with essential scent design principles and ingredient profiles |Nova Canvas | Visual creative generator | Creates fragrance visuals and naming concepts; outputs marketing-ready imagery |Nova Reel | Video generation | Transforms visuals into dynamic campaign videos; supports customization |Amazon Polly | Text-to-speech voice | Provides a French-accented voice for campaign videos |Bedrock Guardrails | Safety and governance | Filters content to block allergens and undesirable topics |Perfumers (on-site) | Human-in-the-loop | Finalize hundreds of unique fragrances per day based on AI-generated concepts |

Key takeaways

The Fragrance Lab is a holistic demonstration of combining conversational AI, AI-assisted product design, and AI-generated marketing at scale within Bedrock.
The architecture emphasizes trait-driven personalization, real-time knowledge access via RAG, and end-to-end asset generation for products and campaigns.
Safety and governance are integral, with Bedrock Guardrails helping to block unwanted topics and ensure compliant customer interactions.
The model stack is modular and adaptable, enabling new categories and variations such as translations and sizing to be added as needed.
The collaboration with Wildlife showcases how AI services can be translated into compelling physical experiences that resonate with audiences in real-world settings.

FAQ

What is The Fragrance Lab built to demonstrate?

It demonstrates an end-to-end, personalized product and marketing-asset pipeline built with Amazon Nova in Bedrock, combining conversational AI, fragrance design, and AI-generated campaigns.
Which components are used to create and market personalized fragrances?

Nova Sonic handles dialogue, Nova Pro with RAG designs fragrances, Nova Canvas creates visuals, Nova Reel makes videos, Polly provides voice, and Guardrails ensure safe interactions, all orchestrated in Bedrock.
How does it accelerate fragrance development?

The system enables hundreds of unique fragrances per day, turning what would normally take hours for a perfumer into a faster AI-assisted process, with human perfumers finalizing the results.
Can this approach be used beyond fragrance?

Yes, the architecture is designed to be replicated across multiple retail and consumer goods categories, including skincare, fashion, food and beverage, home goods, and wellness.