Accelerating your advertising ideation with generative AI – Half 1: From thought to era with the Amazon Nova basis fashions

Advertising groups face rising stress to create participating campaigns shortly whereas sustaining model consistency and inventive high quality. Conventional advertising marketing campaign creation processes usually contain a number of iterations between artistic groups, stakeholders, and exterior businesses, resulting in prolonged timelines and elevated prices.

The appearance and availability of generative fashions (particularly picture and video era ones) has opened the chance to shortly iterate by means of a number of marketing campaign proposals in minutes. Nonetheless, environment friendly marketing campaign creation aided by generative fashions nonetheless requires a excessive degree of expertise and mastery of generative instruments comparable to immediate engineering, parameter fine-tuning, utility of guardrails, and so forth. Enhancing, scripting, and post-production expertise are additionally necessary to create gorgeous tales.

On this put up, the primary of a collection of three, we deal with how you should utilize Amazon Nova to streamline, simplify, and speed up advertising marketing campaign creation by means of generative AI. We present how Bancolombia, one in all Colombia’s largest banks, is experimenting with the Amazon Nova fashions to generate visuals for his or her advertising campaigns.

The problem of contemporary advertising campaigns

The challenges to create participating advertising campaigns prolong far past artistic overhead, impacting companies at operational, monetary, and strategic ranges. The traditional method to advertising marketing campaign creation sometimes includes a posh technique of interactions between inner groups, exterior businesses, and stakeholders—every with their very own priorities, suggestions cycles, and approval processes.

Probably the most important ache factors in conventional advertising marketing campaign creation is the time-intensive nature of the work. From preliminary idea improvement to last asset manufacturing, campaigns usually require weeks and even months to maneuver from ideation to execution. This prolonged timeline may embody a number of rounds of revisions, stakeholder approvals, and asset refinements that may symbolize an overhead within the inner course of. Nevertheless, when speaking in regards to the advertising enterprise, the context during which campaigns are launched is what issues. Exterior elements like latest information, trend developments, new launched merchandise, and season of the yr can affect the person’s acceptance of the campaigns. That’s the reason time is important for advertising, and a delayed launch for the marketing campaign can imply lacking important home windows or permitting rivals to seize viewers consideration first.

The monetary implications of those conventional strategies are equally regarding. Advertising groups incessantly face finances constraints whereas being anticipated to ship refined campaigns throughout a number of channels and supply codecs comparable to social media posts, quick movies, touchdown pages, and extra. The prices related to a marketing campaign creation—together with company charges, manufacturing bills, and the chance price of delayed launch—can shortly escalate. In line with a Gartner survey, in 2024, corporations diminished their advertising finances from 9.9% to 7.7% of their complete finances, additional constraining the sources and time allotted for creating advertising campaigns. This finances crunch is exactly why AI instruments have gotten important reasonably than non-compulsory for advertising groups.

Past time and value concerns, advertising groups battle with a elementary rigidity. They have to preserve constant model identification whereas creating contemporary, participating content material on the tempo {the marketplace} calls for. This problem will get much more difficult by content material personalization throughout numerous programs and codecs, every with their very own technical necessities and finest practices. In response, we’re seeing corporations fully rethink how they allocate sources. Many are shifting towards paid media and reducing again on legacy applied sciences, decreasing headcount, and scaling again company relationships. This creates an attention-grabbing scenario during which AI instruments are each inflicting finances reductions (by making groups extra environment friendly) and serving to groups survive those self same finances cuts.

The true-world affect is important. Advertising groups utilizing generative AI can now produce in hours what used to take days or perhaps weeks. These instruments aren’t simply creating effectivity—they’re enabling completely new approaches to content material creation. This technological shift has created excellent circumstances for implementing superior generative fashions that may produce high-quality photos, movies, and different visible property at a scale beforehand inconceivable with conventional strategies.

Amazon Nova household of basis fashions

Amazon Nova is a household of basis fashions (FMs), obtainable utilizing APIs by means of Amazon Bedrock, created by Amazon, and differentiated by its nice price-performance fee.

The household consists of the next subfamilies:

• Amazon Nova Understanding Fashions – Understanding fashions that settle for textual content, picture, and video inputs and generate textual content output. They supply a broad collection of functionality, accuracy, pace, and value operation factors. Composed of 4 fashions: Amazon Nova Micro, Amazon Nova Lite, Amazon Nova Professional, Amazon Nova Premier.
• Amazon Nova Artistic Content material Technology Fashions – Artistic content material era fashions that settle for textual content and picture inputs and produce picture or video outputs. Built-in by two fashions: Amazon Nova Canvas (picture era) and Amazon Nova Reel (video era).
• Amazon Nova Speech-to-Speech Fashions – Consists of just one mannequin, Amazon Nova Sonic, a speech-to-speech mannequin that accepts speech as enter and generates speech and textual content as output. The mannequin is designed to ship real-time, humanlike voice conversations with contextual richness.

Picture era with Amazon Nova Canvas

With Amazon Nova Canvas you possibly can generate life like, studio-quality photos through the use of textual content prompts. Amazon Nova Canvas is able to producing photos as much as 2K x 2K. Amazon Nova Canvas can also be able to picture modifying present photos utilizing textual content prompts to information the version. Check with the Visible information to Amazon Nova Canvas to discover the chances provided by the Nova Canvas mannequin. To programmatically generate a picture, enter the next code:

import boto3
import base64
import io
import json
from PIL import Picture
# Initialize the Bedrock shopper
bedrock_runtime = boto3.shopper(
    service_name="bedrock-runtime",
    region_name="us-east-1"  # Specify your area
)
# Outline your immediate for picture era
immediate = "A serene mountain panorama with a lake at dawn, photorealistic type"
# Create the request payload for Amazon Nova Canvas
request_payload = {
    "taskType": "TEXT_IMAGE",
    "textToImageParams": {"textual content": immediate},
    "imageGenerationConfig": {
           "cfgScale": 7,
           "seed": 42,
           "numberOfImages": 1,
           "width": 1024,
           "top": 1024
    }
}
# Name the Bedrock mannequin
response = bedrock_runtime.invoke_model(
    modelId="amazon.nova-canvas-v1:0", 
    physique=json.dumps(request_payload)
)
# Course of the response
response_body = json.hundreds(response["body"].learn())
image_base64 = response_body["images"][0]
# Convert base64 to picture
image_data = base64.b64decode(image_base64)
picture = Picture.open(io.BytesIO(image_data))
# Save the picture
picture.save("generated_image.png")
print("Picture generated and saved as 'generated_image.png'")

Discover extra Amazon Nova examples within the Amazon Nova samples GitHub repository.

You possibly can study immediate engineering for Amazon Nova Canvas and Amazon Nova Reel at Picture and video immediate engineering for Amazon Nova Canvas and Amazon Nova Reel within the AWS Synthetic Intelligence Weblog.

Speed up the creation of visible property for advertising campaigns utilizing Amazon Nova

To streamline the era of promoting marketing campaign visible property we suggest an automatic course of, aided by generative AI, to transition from marketing campaign thought to a set of visible property for the marketing campaign. Our proposed course of is described as follows:

1. You present an outline of your marketing campaign
2. The system robotically retrieves earlier marketing campaign photos associated to your present marketing campaign, and you may choose people who higher represents the concepts to your new marketing campaign
3. The system will generate a reference immediate (utilizing Amazon Nova Professional) to generate your marketing campaign’s photos utilizing Amazon Nova Canvas. You possibly can edit this immediate.
4. The system will generate as much as 5 photos to your marketing campaign.

We now discover a few of the key ideas of this resolution.

Reference picture advice

After offering the outline of our picture, we seek for photos associated to our marketing campaign inside a picture database. Such a search returns photos intently associated to the marketing campaign matter that have been utilized in earlier campaigns and have confirmed profitable prior to now. The person can choose a few of the reference photos to information the creation of our new visible property. This method is the subject of debate of the following installment on this collection. We encourage you to learn it to achieve a deeper understanding of how we created an environment friendly search engine utilizing multimodal embeddings fashions and vector databases.

Creating good picture era prompts utilizing giant language fashions (LLMs)

Though Amazon Nova artistic fashions have some understanding of languages comparable to Spanish or German, English stays the one totally supported language. Due to this fact, customers ought to craft their prompts in English to maximise the capabilities of Amazon Nova artistic fashions. Furthermore, prompting artistic fashions successfully requires a unique method than when working with understanding fashions. Even skilled immediate engineers may have time to adapt their strategies to completely unlock a artistic mannequin’s potential. For newcomers to LLM-based content material era, this studying curve may be notably steep.

To handle these challenges, we use a method referred to as metaprompting. This method includes instructing one LLM to generate efficient prompts for different FMs. For instance, we will use Amazon Nova Professional to craft high-quality prompts for Amazon Nova Canvas to generate photos.

The next immediate template demonstrates the right way to instruct Amazon Nova Professional to generate optimized text-to-image prompts based mostly on a easy picture description (highlighted in daring):

You're a graphics designer named Joe that focuses on creating visualizations aided by text-to-image basis fashions.

Your colleagues come to you every time they wish to craft environment friendly prompts for creating photos with text-to-image basis fashions comparable to Nova Canvas.
You at all times reply to your colleagues requests with a really environment friendly immediate for creating nice visualizations utilizing text-to-image basis fashions.

These are some guidelines you'll observe when interacting along with your colleagues:

* Your colleagues will talk about their concepts of their native languages, so please be versatile.
* Your solutions will at all times be in English whatever the language your colleague used to speak.
* Your immediate ought to be at most 512 characters. You're inspired to make use of all of them.
* Don't give particulars about or decision of the pictures within the immediate you'll generate.
* You'll at all times say out loud what you're considering
* You at all times cause solely as soon as earlier than making a immediate
* It doesn't matter what you at all times present a immediate to your colleagues
* You'll create just one immediate
* If supplied with reference picture descriptions (can be in between  XML tags) rigorously steadiness the contributions of the campaigns description with the reference photos to create the immediate
* By no means counsel so as to add textual content to the pictures

Listed below are some pointers you at all times observe when crafting efficient picture prompts:

* Begin with a Clear Imaginative and prescient: Have a transparent thought of the picture you need the AI to generate, picturing the scene or idea in your thoughts intimately.
* Select Your Topic: Clearly state the primary topic of your picture, making certain it’s prominently talked about within the immediate.
* Set the Scene: Describe the setting or background, together with the setting, time of day, or particular location.
* Specify Lighting and Environment: Use descriptive phrases for lighting and temper, like “bathed in golden hour mild” or “mystical ambiance”.
* Incorporate Particulars and Textures: Enrich your immediate with descriptions of textures, colours, or particular objects so as to add depth.
* Keep away from utilizing negation phrases within the immediate creation. The usage of negations is prohibited since this normally yields negative effects within the generated photos.
* Be Aware of Size and Readability: Efficient prompts are typically detailed however not overly lengthy, offering key visible options, types, feelings or different descriptive components.
* Particular tokens may be added to supply higher-level steerage: Like “photorealistic”, “cinematic lighting” and so on. These act like key phrases for the mannequin.
* Logically ordering immediate components and utilizing punctuation to point relationships: For instance, commas to separate unbiased clauses or colons to guide into an outline.

Create a immediate for: A credit score marketing campaign for small espresso producers within the mountains of Colombia.

Output solely the immediate

The reply is:

A vibrant, photorealistic scene showcasing small espresso producers within the lush, 
inexperienced mountains of Colombia, bathed in golden hour mild. Farmers, dressed 
in conventional apparel, are engaged within the espresso harvesting course of, surrounded 
by espresso crops and picturesque landscapes.The ambiance is heat and welcoming, 
with a deal with group and sustainable practices. Embrace components like 
rustic wood instruments, ripe espresso cherries, and the majestic Andean backdrop. 
Keep away from fashionable know-how and concrete components.

As you possibly can see, we get a immediate that’s aligned with the necessities of Amazon Nova Canvas, and we get a high-quality picture.

Out metaprompt template has 4 sections:

1. We set the persona and the high-level activity (environment friendly immediate creation).
2. We set the foundations we anticipate to be adopted when writing the immediate. For instance, we offer the character limits and instruct the mannequin to reply solely in English.
3. We set pointers that describe how we would like the immediate to be constructed however usually are not necessary. There are additionally some hints about key phrases used within the last mannequin, so the metaprompt encourages the mannequin to make use of them in its output.
4. We establish the ultimate output.

If you wish to discover extra metaprompting concepts, the weblog put up Enhance Amazon Nova migration efficiency with data-aware immediate optimization has good insights. For finest practices for writing prompts for Amazon Nova understanding fashions, discuss with the Amazon Nova Consumer Information.

Answer structure

We now suggest a reference structure to implement the concepts mentioned above. The structure is as follows:

1. Amazon Bedrock is used to invoke the Amazon Nova Professional and Amazon Nova Canvas fashions utilizing a standard API.
2. Amazon Lambda features are used to create metaprompts, generate photos, retrieve associated photos and retailer the marketing campaign in a database.
3. The marketing campaign’s data is saved in an Amazon DynamoDB database.
4. APIs are managed utilizing Amazon API Gateway.
5. Amazon Cognito is used to handle the customers of the applying.

The next diagram reveals the structure.

How Bancolombia is utilizing Amazon Nova to streamline their advertising marketing campaign property era

Bancolombia, one in all Colombia’s main banks has been experimenting with this method for over a yr now. Often, strategic advertising in Bancolombia begins with a quick that units out the objectives, target market and primary messages. An in depth temporary, usually containing visible examples and knowledge evaluation, can assist streamline the workflow between the groups and stakeholders.

“There are sometimes many conferences and rounds of modifications wanted between making the temporary and getting all of the stakeholders aligned”, stated Juan Pablo Duque, Advertising Scientist Lead at Bancolombia. “That is the place we realized we may usher in new applied sciences. The saying ‘an image is value a thousand phrases’ could be very true. It could be nice if advertising analysts may use generative AI to make visible references. Moderately than simply utilizing textual content, they could draw a transparent illustration of what they’re considering.” 

“We began to experiment, and after each iteration, we discovered extra choices. For instance, what if we checked out data from earlier campaigns? Many earlier initiatives had the identical objectives and the way they carried out can be utilized as helpful data. Utilizing visuals along with confirmed marketing campaign components would give the artistic crew a stronger basis which might assist them save time within the early phases of making concepts.

As generative fashions hold enhancing, they might cowl the complete design course of by producing photos and adjusting them for various digital makes use of. This implies producing content material following our platform’s pointers. With this workflow, publishing may be achieved sooner, groups can react to present developments, enhance on content material that isn’t doing effectively and provide messages which are extra constant.” 

Implementation finest practices

On this part, we suggest a set of finest practices for readers meaning to automate the era of promoting campaigns.The next are some technical concerns and pointers:

1. Modular structure – Implement the answer utilizing a modular method, separating the picture search engine and visible era parts. This enables for simple upkeep and future enhancements.
2. Serverless first – Use serverless applied sciences comparable to AWS Lambda and Amazon API Gateway to scale back operational overhead and enhance scalability.
3. Mannequin model management – Hold observe of the particular variations of Amazon Nova fashions utilized in your implementation. This helps reproducibility and helps in troubleshooting.

For safety finest practices, think about these pointers:

1. Least privilege entry – Use AWS Identification and Entry Administration (IAM) roles with the precept of least privilege for all parts, particularly when accessing Amazon Bedrock and different AWS providers.
2. Encryption in transit and at relaxation – Be sure all knowledge, together with marketing campaign descriptions and generated photos, is encrypted each in transit and at relaxation utilizing AWS Key Administration Service (AWS KMS).
3. Enter validation – Implement strict enter validation for all person inputs to assist stop potential safety vulnerabilities, particularly within the marketing campaign description and immediate modifying phases.
4. Safe API entry – Use Amazon Cognito for person authentication and authorization to safe entry to your API endpoints.

To optimize efficiency, observe these pointers:

1. Caching technique – Implement a caching layer utilizing Amazon ElastiCache to retailer incessantly accessed knowledge, comparable to standard marketing campaign descriptions or generally used reference photos.
2. Asynchronous processing – For picture era duties that may take longer, implement asynchronous processing to enhance person expertise and system responsiveness.
3. Auto scaling – Configure auto scaling to your Lambda features and API Gateway to deal with various hundreds effectively.

For error dealing with and monitoring, observe these pointers:

1. Complete logging – Implement detailed logging utilizing Amazon CloudWatch Logs to seize system occasions, person actions, and mannequin interactions.
2. Automated alerts – Arrange CloudWatch alarms to observe key metrics and ship notifications for anomalies or errors.
3. Sleek degradation – Design the system to gracefully deal with failures, comparable to momentary unavailability of the Amazon Nova fashions, by implementing acceptable fallback mechanisms.
4. Common audits – Conduct periodic audits of your generated content material to confirm compliance with model pointers and moral requirements.

By following these finest practices, you possibly can preserve a dependable, safe, and environment friendly implementation of your generative AI advertising marketing campaign resolution. Common monitoring and upkeep can assist you establish areas for enchancment and hold your system optimized as your marketing campaign wants evolve.

Conclusion

On this put up, we demonstrated the right way to use the Amazon Nova household of FMs to revolutionize advertising marketing campaign creation by means of an automatic, AI-driven method. By combining Amazon Nova Professional for clever immediate era and Amazon Nova Canvas for high-quality picture creation, we’ve proven how advertising groups can dramatically scale back the time and sources required to provide compelling visible property whereas sustaining model consistency and inventive high quality.Trying forward, this resolution may be prolonged to help extra use instances comparable to:

• Integration with advertising automation providers
• Superior personalization utilizing buyer insights
• Multi-channel marketing campaign asset era
• A/B testing automation for visible content material

Within the subsequent installment of this collection, we are going to handle the right way to incorporate data from earlier marketing campaign photos to enhance the artistic course of.

We additionally encourage you to discover the capabilities of the Amazon Nova household of fashions to your advertising workflows. You may get began by deploying the answer publicly obtainable within the generative AI ML latam samples GitHub repository.

In regards to the authors

David Laredo is a Prototyping Architect at AWS, the place he helps prospects uncover the artwork of the doable by means of disruptive applied sciences and speedy prototyping strategies. He’s keen about AI/ML and generative AI, for which he writes weblog posts and participates in public talking periods throughout LATAM. He at the moment leads the AI/ML specialists group in LATAM.

Josué Martínez is Sr Options Architect with +25 years of expertise in IT. Final 10 in Cloud Applied sciences with deal with AI/ML options. Josue loves nature an spending time to its household figuring out magic locations. Nonetheless coding for enjoyable.

Dr. José Gardiazabal is a Options Architect at AWS, centered on Healthcare and Life Sciences. He’s keen about studying new applied sciences, and discovering out the right way to assist prospects with them. He’s additionally working with prospects utilizing IoT throughout Latin America.

Arturo Minor is a prototyping options architect based mostly in Mexico Metropolis. He enjoys the outside and touring, and is serious about laptop science, languages, and know-how. He works at AWS serving to shoppers innovate, develop, and visualize new options.

Jorge Lopez is a Sr Options Architect at AWS with over 10 years of expertise in know-how. He’s a technical chief in AI/ML at AWS.

Juan Pablo Duque is a Advertising Science Lead at Bancolombia, the place he merges science and advertising to drive effectivity and effectiveness. He transforms advanced analytics into compelling narratives. Captivated with GenAI in MarTech, he writes informative weblog posts. He leads knowledge scientists devoted to reshaping the advertising panorama and defining new methods to measure.