AI Agents for Advertising: A Practical Guide to Creative Asset Generation

How to build an AI agent system using Strands Agents SDK and Amazon Bedrock to automate advertising creative generation — from copywriting to image and video production.

zhuermu · · 15 min
AI AgentAdvertisingAIGCAWSStrands AgentsAmazon Nova

中文版 / Chinese Version: This article was originally published on the AWS China Blog. Read the original →

1. Introduction

The advertising industry is undergoing a thrilling transformation. AI agent tools are poised to completely reshape how marketing campaigns are created, digital ads are produced, and marketing assets are localized. While these innovations promise massive operational efficiency gains for businesses, the technology is evolving so rapidly that many organizations are still figuring out which AI tools to adopt and how to effectively integrate them into existing workflows.

Traditional display ad and video ad creative development typically takes weeks or months to produce multiple variants in different formats and sizes for various placements. The speed of asset creation directly impacts campaign launch timelines, and advertisers need the ability to respond to market changes faster than ever. Today’s agentic AI technology can dramatically accelerate the creative production process — which means advertising professionals need to start exploring and experimenting with these new technologies early to fully harness them and ride the wave of AI-driven innovation.

Once an organization has a large volume of AI-generated creative assets, completing the intelligent advertising loop requires two additional key components: an ad placement assistant that precisely distributes assets across various channels, and a dynamic creative optimization (DCO) assistant that monitors asset performance in real time and continuously optimizes delivery outcomes. Only by organically linking creative generation, ad placement, and performance optimization can you maximize AI’s efficiency gains at every node and truly achieve an end-to-end intelligent advertising system.

This article focuses on the AI agent-driven creative asset generation phase. We will dive deep into how to use AI agent technology to streamline the creative production process, walking through real-world examples and sample code to demonstrate how to build a fully functional creative design agent.

2. Key Use Cases for AI Agents in Advertising and Marketing

Based on the emerging value chain in advertising and marketing, we have identified the following AI agent use cases:

Campaign creative concept and brief development. Creative departments at advertising agencies want to leverage AI agents to enhance and accelerate the creative development process. With multimodal LLMs (such as Amazon Nova Pro), creative teams can quickly extract insights from brand guidelines, product descriptions, past campaigns, and existing marketing materials — generating campaign briefs in minutes rather than days.

Display ad and video ad generation. Traditional creative production teams need to produce multiple variants of the same ad, a process that typically takes weeks. AI agent-based approaches can combine LLMs, image generation models, and video generation models to produce assets at lower cost and higher speed.

Intelligent ad placement. Generative AI combined with agent technology can power intelligent ad placement assistants that analyze historical delivery data, user behavior patterns, competitor placement strategies, and other multi-dimensional information to automatically generate placement plan recommendations.

Ad performance analysis and optimization. Dynamic creative optimization assistants built with AI agents can aggregate data from different delivery platforms in real time, leveraging the reasoning capabilities of LLMs to automatically identify high-performing and underperforming assets along with their distinguishing characteristics.

This article focuses specifically on creative development and asset generation for video advertising.

3. The Evolution of Advertising Creative Generation Tools

3.1 From Fragmented Tools to Integrated Solutions

The market today is filled with specialized tools for specific tasks — chatbots paired with precise prompts for developing ad concepts, Stable Diffusion or Flux models for image generation, and Kling (by Kuaishou) for video production. However, very few users can proficiently navigate such a diverse toolchain.

3.2 The Evolution of AI Image Generation Tools

Phase 1: High Technical Barrier (2023 and earlier) — Early AI image generation relied heavily on Stable Diffusion models combined with various LoRA models. Users needed to perform complex operations in the SD-WebUI interface, requiring significant technical expertise.

Phase 2: Professional Specialization (2024) — ComfyUI gained popularity, and open-source models evolved to the Flux series. However, the user base remained limited to AI enthusiasts and technical power users.

Phase 3: The AI Agent Revolution (2025) — The launch of Manus introduced the general public to the concept of AI agents. Design agent tools like Lovart brought a revolutionary conversational interaction paradigm: users simply describe their requirements and upload relevant images, and the system generates various image or video content automatically.

4. Building a Creative Asset Generation Agent with Strands Agents

4.1 Introduction to Strands Agents

Strands Agents is an open-source, lightweight agent development SDK that takes a model-driven approach to building and running AI agents. It enables developers to create agents with autonomous planning, reasoning, and tool selection capabilities in just a few lines of code.

4.2 Overall Architecture Design

The system adopts a Strands-based layered multi-agent architecture:

  • Separation of concerns: Copywriting, image generation, and video generation are three independent modules, each handled by a specialized sub-agent.
  • Extensibility: The “Agents as Tools” pattern allows new capabilities to be added without modifying the orchestration logic.
  • Fault tolerance: The main agent (Ads Agent) handles task orchestration and error recovery, isolating failures in individual sub-agents.
  • Flexibility: The main agent intelligently determines which sub-agents to invoke based on the user’s request.

4.3 Workflow Design

The agent operates through three distinct phases:

  • Requirement Understanding Phase: The Ads Agent analyzes the user’s input, extracts key information such as product type, target audience, brand tone, and desired output formats.
  • Task Planning Phase: The agent generates a structured execution plan, breaking the request into discrete tasks assigned to the appropriate sub-agents.
  • Parallel Execution Phase: For tasks with no dependencies between them, the agent can invoke multiple sub-agents in parallel to maximize throughput.

4.4 Technical Implementation

Here is how the main Ads Agent is constructed using Strands Agents:

from strands import Agent, tool

def create_ads_agent():
    # Define available tools including three specialist agents
    all_tools = [
        image_generator,   # Image generation agent tool
        video_generator,   # Video generation agent tool
        copy_writer,       # Copywriting agent tool
    ]
    ads_agent = Agent(
        model=init_model(),
        system_prompt="<Ads Agent System Prompt>",
        tools=all_tools,
    )
    return ads_agent

Each sub-agent is wrapped as a tool using the @tool decorator, following the “Agents as Tools” pattern.

Image Generator Agent

The image generation sub-agent is responsible for producing visual assets. It takes a creative brief and generates images using models like Amazon Nova Canvas or Stability AI:

@tool
def image_generator(prompt: str, aspect_ratio: str = "16:9") -> str:
    """Generate advertising images based on the creative brief.

    Args:
        prompt: Detailed description of the desired image
        aspect_ratio: Output aspect ratio (e.g., "16:9", "1:1", "9:16")
    """
    image_agent = Agent(
        model=init_model(),
        system_prompt="""You are a professional advertising image designer.
        Generate high-quality advertising visuals based on the brief.
        Consider brand guidelines, color schemes, and visual hierarchy.
        Use the image generation tool to create the final output.""",
        tools=[generate_image_tool],
    )
    result = image_agent(f"Generate an ad image: {prompt}, aspect ratio: {aspect_ratio}")
    return str(result)

Copywriter Agent

The copywriting sub-agent generates advertising copy — headlines, taglines, body text, and calls to action:

@tool
def copy_writer(product_info: str, target_audience: str, tone: str = "professional") -> str:
    """Generate advertising copy for the specified product.

    Args:
        product_info: Product description and key selling points
        target_audience: Target audience demographics and interests
        tone: Desired copy tone (e.g., "professional", "playful", "urgent")
    """
    copy_agent = Agent(
        model=init_model(),
        system_prompt="""You are a senior advertising copywriter.
        Write compelling ad copy that resonates with the target audience.
        Include headlines, taglines, body copy, and CTAs.
        Ensure copy aligns with brand voice and campaign objectives.""",
        tools=[],
    )
    result = copy_agent(
        f"Write ad copy for: {product_info}. "
        f"Target audience: {target_audience}. Tone: {tone}"
    )
    return str(result)

Video Generator Agent

The video generation sub-agent handles creating short-form video ads from images and prompts using models like Amazon Nova Reel:

@tool
def video_generator(prompt: str, image_url: str = None, duration: int = 6) -> str:
    """Generate advertising video based on the creative brief.

    Args:
        prompt: Detailed description of the desired video content and motion
        image_url: Optional reference image to use as the first frame
        duration: Video duration in seconds
    """
    video_agent = Agent(
        model=init_model(),
        system_prompt="""You are a professional video advertising producer.
        Create compelling short-form video ads.
        Consider visual storytelling, pacing, and viewer engagement.
        Use the video generation tool to produce the final output.""",
        tools=[generate_video_tool],
    )
    result = video_agent(
        f"Generate an ad video: {prompt}. Duration: {duration}s. "
        f"Reference image: {image_url if image_url else 'None'}"
    )
    return str(result)

Running the Agent

Here is how the complete system is invoked:

if __name__ == "__main__":
    ads_agent = create_ads_agent()
    ads_agent("Create a one-shot ads video for a camping tent")

When executed, the Ads Agent automatically:

  1. Analyzes the request — identifies the product (camping tent), the desired output (video ad), and the style (one-shot).
  2. Creates a plan — determines it needs copywriting for the ad text, an image for the key visual, and a video as the final deliverable.
  3. Invokes the Copywriter Agent — generates a compelling headline and tagline for the camping tent.
  4. Invokes the Image Generator Agent — creates a hero image of the camping tent in a scenic outdoor setting.
  5. Invokes the Video Generator Agent — produces a short video ad using the generated image as the first frame, with camera motion and text overlays.
  6. Returns the results — delivers all generated assets (copy, image, video) back to the user with download links.

5. Enterprise-Grade Solution

5.1 Real-World Business Impact

Organizations deploying AI agent-based creative generation have seen remarkable results:

  • Dramatic efficiency gains — Average creative asset production time reduced by 85%, with monthly asset output volume increasing 5x.
  • Clear conversion rate improvements — AI-generated video ads achieved 3x click-through rate growth compared to non-AI videos, delivering a 21% improvement in ROAS (Return on Ad Spend).

5.2 Enterprise Features

A production-ready enterprise solution extends beyond the core agent architecture to include:

  • Multi-model support — Integration with multiple image generation models (Amazon Nova Canvas, Stability AI, Flux) and video generation models (Amazon Nova Reel, Kling) to give creative teams maximum flexibility.
  • Project management — Organize campaigns, briefs, and generated assets into structured projects with version history and approval workflows.
  • Asset management — Centralized library for all generated creative assets with tagging, search, and filtering capabilities.
  • Multi-user access — Role-based access control with team collaboration features and audit trails.

5.3 Technical Architecture

The enterprise solution is deployed on the following AWS infrastructure:

  • Amazon ECS — Container service hosting the application, providing auto-scaling and high availability.
  • Application Load Balancer (ALB) — Provides external access with SSL termination and request routing.
  • Amazon S3 — Stores all generated creative assets (images, videos, documents) produced during user sessions.
  • Amazon DynamoDB — Persists project definitions, campaign metadata, and asset records.
  • Amazon Cognito — Handles user authentication and authorization with support for enterprise SSO integration.

This architecture provides the scalability, security, and reliability required for enterprise advertising workflows while keeping operational overhead minimal through managed AWS services.

6. Conclusion

AI agents are reshaping the creative production process across the advertising and marketing industry. Through an agent-based architecture, complex AI tools can be unified into an intuitive conversational interface — lowering the barrier to entry while dramatically increasing creative output.

Key takeaways from this guide:

  • The multi-agent pattern works: Separating copywriting, image generation, and video generation into specialized sub-agents — orchestrated by a main agent — provides clean separation of concerns with maximum flexibility.
  • Strands Agents simplifies development: The SDK’s model-driven approach and “Agents as Tools” pattern make it straightforward to build sophisticated multi-agent systems.
  • Enterprise readiness requires more than just agents: Production deployments need project management, asset management, multi-user support, and robust cloud infrastructure.

Looking ahead, the future directions for AI in advertising include:

  • Stronger multimodal understanding — Agents that can analyze existing brand assets, competitor campaigns, and market trends to generate more contextually relevant creative.
  • Real-time collaboration — Multiple team members working with the same agent simultaneously, with shared context and coordinated outputs.
  • Intelligent creative recommendations — Proactive suggestions based on historical performance data and emerging creative trends.
  • End-to-end marketing automation — Fully integrated pipelines from creative generation through placement optimization to performance analysis, all orchestrated by AI agents.