ViGenAiR is an open-source tool that leverages Google's multimodal Generative AI, Gemini, to transform existing video ads into shorter versions tailored for different formats and audiences. This innovation addresses the critical Creative pillar in advertising, which can unlock up to 50% ROI according to the Google Marketing team.
ViGenAiR focuses on generating high-quality video assets in various durations and formats, addressing a significant pain point for advertisers. This solution ensures that video ads maintain coherence and adhere to best practices, enhancing their effectiveness.
Benefits
- Inventory: Supports horizontal, vertical, and square video assets, expanding reach across all Google-owned platforms.
- Campaigns: Produces shorter, compelling video ads ideal for social media and awareness campaigns.
- Creative Excellence: Ensures videos are coherent and follow best practices, including upcoming creative direction rules.
- User Control: Allows users to guide the model via prompts or manual scene selection.
- Performance: Built-in A/B testing to identify the best-performing variants.
How ViGenAiR Works
ViGenAiR is an Angular Progressive **Web App hosted **on Google Apps Script, requiring Google account authentication. Backend services are hosted on Cloud Functions 2nd gen, triggered via Cloud Storage, ensuring a separation of concerns between frontend and backend.
Using Gemini on Vertex AI, ViGenAiR analyzes video content, splitting it into coherent audio/video segments. These segments are recombined to create shorter variants that maintain the original ad's storyline or introduce new ones, adhering to Google's best practices.
Limitations
- Not suitable for all video types.
- Users cannot delete previously analyzed videos via the UI.
- Current audio tech cannot differentiate between voice-over and singing.
- Generated segments might not always follow user prompts or desired durations.
- Audio overlay settings are only applied to fully rendered variants.
- Adherence to creative rules requires additional prompt instructions or a distilled language model based on brand-specific guidelines.
For more details, please visit the GitHub repository for the tool.
