What is ViGenAiR

ViGenAiR is a project that leverages generative AI to recraft video ads, enhancing their effectiveness and engagement. This tool addresses the challenges of creating compelling video content by automating and optimizing the ad creation process. It allows users to process videos of any length or size, even beyond traditional limits, and includes features like manual adjustment of the Smart Framing crop area to better capture points of interest. With continuous updates and improvements, ViGenAiR is worth exploring for anyone looking to innovate in video advertising.

Project Overview

Key Features

Overview

ViGenAiR is a project that leverages multimodal Generative AI models on Google Cloud Platform (GCP) to transform long-form video ads into shorter variants in multiple ad formats. It generates video, image, and text assets for Demand Gen and YouTube video campaigns.

Key Features

• Video Transformation: Automatically splits long-form video ads into coherent audio/video segments to generate shorter variants.
• Ad Formats: Supports horizontal, vertical, and square video formats.
• User Control: Users can steer the model towards specific aspects or exclude certain elements of the input video.
• Creative Excellence: Generates videos that follow Google’s best practices for effective video ads, including dynamic framing and adherence to creative rules.
• Performance: Built-in A/B testing to identify the best variants for target audiences.
• Demand Gen Assets: Generates text and image assets for Demand Gen campaigns using multimodal models.

How it Works

• Frontend: An Angular Progressive Web App hosted on Google Apps Script, accessible via a web app deployment.
• Backend: Cloud Functions 2nd gen on GCP, triggered via Cloud Storage (GCS).
• Video Processing: Extracts video information, separates background music and voice-over, transcribes voice-over, and analyzes visual shots using Cloud Video AI API.
• Variants Generation: Uses Vertex AI to generate variants based on user prompts and target durations.
• Rendering: Renders final videos in desired formats and settings via the Combiner service Cloud Function.

Benefits

• Inventory Utilization: Taps into all Google-owned sources of inventory.
• Campaign Optimization: Ideal for social and awareness/consideration campaigns.
• Creative Control: Allows users to focus on or exclude specific topics or elements.
• Performance Improvement: Automatically identifies the best variants through A/B testing.

Requirements and Deployment

• Requires a Google account, GCP project, and specific roles (Vertex AI User, Storage Object User, etc.).
• Deployment involves setting up GCS buckets, Cloud Functions, and Apps Script deployments.

Pricing and Quotas

• Priced based on usage of Google Cloud and Workspace services, including Cloud Storage, Cloud Functions, and Vertex AI models.
• Detailed pricing calculator example available for cost estimation.

Contribution

• Follows standard contributing guidelines with additional steps for building and deploying GCP components and the Angular UI locally.

Real-World Applications

Campaign Optimization

• Shorter Video Ads: Use ViGenAiR to transform long-form video ads into shorter, more compelling variants suitable for social media and awareness/consideration campaigns. This helps in capturing audience attention more effectively.
  • Upload a long-form video ad.
  • Generate shorter variants using the UI’s variant generation features.
  • Select and render the most engaging short videos.

Multi-Format Advertising

• Horizontal, Vertical, and Square Formats: Generate video ads in various formats to tap into different inventory sources across Google-owned platforms.
  • Upload a video ad.
  • Use the smart framing feature to dynamically frame the most prominent elements in vertical and square formats.
  • Render and preview the videos in all formats.

User-Controlled Creative Generation

• Customized Video Variants: Steer the AI model to focus on or exclude specific aspects of the video, such as certain entities or topics, to align with your creative vision.
  • Provide an optional prompt to guide the variant generation.
  • Use the “Exclude from video variants” checkbox to exclude unwanted elements.
  • Adjust the target duration and generate variants accordingly. ``

Performance Enhancement

• A/B Testing and Analytics: Utilize built-in A/B testing to identify the best-performing video variants tailored to your target audience.
  • Generate multiple variants with different settings.
  • Add variants to the render queue and render them.
  • Analyze the performance of each variant using the score and reasoning provided by the LLM.

Exploring and Benefiting from the Repository

Getting Started

• Follow the Get Started section to set up your environment, install necessary tools, and deploy the ViGenAiR application.
  • Install Node.js, npm, clasp, and gcloud CLI.
  • Clone the repository and navigate to the source code directory.
  • Run npm start to deploy GCP components and the UI.

Contributing to ViGenAiR

• Contribute to the project by building and deploying GCP components and the Angular UI locally.
  • Follow the steps in the How to Contribute section to build and deploy GCP components.
  • Build and serve

Conclusion

Key Points

• Automated Video Ad Transformation: ViGenAiR uses multimodal Generative AI to transform long-form video ads into shorter, multiple-format variants, targeting different audiences.
• Creative Excellence: Generates coherent videos that follow Google’s best practices, including dynamic framing and adherence to creative guidelines.
• User Control and Performance: Allows users to steer the model towards desired outcomes and includes built-in A/B testing for identifying best variants.
• Inventory and Campaigns: Enables advertisers to tap into all Google-owned inventory sources with shorter, compelling video ads suitable for social and awareness campaigns.
• Cost-Effective: Pricing varies but is generally cost-effective, especially with the use of Gemini 1.5 Flash models.

Future Potential

• Enhanced Creative Capabilities: Potential to integrate more advanced creative direction rules, such as camera angles and scene cutting, to further enhance video quality.
• Improved User Experience: Ongoing updates to improve the UI, such as better audio analysis and user control over segment selection and rendering settings.
• Scalability: Ability to process videos of any length or size, beyond current API limits, making it a robust solution for large-scale advertising campaigns.
• Community Contributions: Open to contributions, allowing the community to build upon and enhance the project further.

Limitations and Areas for Improvement

• Video Compatibility: May not work well with all types of videos; requires manual adjustments for certain video characteristics.
• Audio Analysis: Current tech cannot differentiate between voice-over and singing voices; requires user intervention.
• Adherence to Guidelines: Variants may not always follow user prompts or desired durations; requires review and potential modification.

Overall, ViGenAiR has significant potential to revolutionize video ad creation by leveraging AI to generate high-quality, targeted video assets efficiently and cost-effectively.

For further insights and to explore the project further, check out the original google-marketing-solutions/vigenair repository.

Attributions

Content derived from the google-marketing-solutions/vigenair repository on GitHub. Original materials are licensed under their respective terms.