ViGenAiR

ViGenAiR - Recrafting Video Ads with Generative AI

Disclaimer: This is not an official Google product.

> Note: This solution uses AI to edit or generate assets for your ads. Whether your content requires labeling depends on the nature of your assets, where your ads serve, and specific legal obligations that apply to you. In some circumstances Google may automatically apply a label. Please consult with your legal team to determine if/when a label is required for your ads. You can add labels in Google’s Ads products using the AI label setting.

Overview •
Get started •
What it solves •
How it works •
How to Contribute

Updates

Update to the latest version by running npm run update-app after pulling the latest changes from the repository via git pull --rebase --autostash; you would need to redploy the *UI* for features marked as frontend, and *GCP components* for features marked as backend.

* [December 2025]
  * frontend + backend: Added support for 3:4 and 4:3 aspect ratios for video previews and rendering.
  * frontend: Added an option to only change the aspect ratio of the video without shortening it.
  * frontend: Added a Prompt library to easily select pre-defined prompts for variant generation.
  * frontend + backend: Added functionality to edit video transcriptions (VTT) and synchronize data.
  * backend: Added functionality to store approval status of rendered variants.
  * backend: Updated generation prompts to support full-length format conversion.
  * frontend + backend: Added Blanking Fill effect for cases where cropping is not viable.
  * frontend + backend: Added automatic conversion of .mov (QuickTime) files to .mp4 format for full browser compatibility and proper video processing. The original .mov file is replaced with the converted .mp4 version. Improved error handling to prevent infinite loading states and provide clear user feedback when conversion fails.
* [July 2025]
  * frontend + backend: Added support for Gemini 2.5 models along with a few fixes and enhancements to the variants generation prompts.
* [May 2025]
  * Significantly reduced the number of deployment steps by utilising Cloud Shell to deploy both backend and frontend components. Read more here.
* [April 2025]
  * frontend + backend: Added functionality to cut segments by adding *split markers* and re-running the extraction process. Read more here.
* [February 2025]
  * frontend: You can now choose objective-specific ABCDs (Awareness, Consideration, Action, or Shorts) in the *Advanced settings* section of variants generation. Read more here.
* [January 2025] Happy New Year!
  * frontend: You can now input your own guidelines for evaluation and scoring of generated variants using the *Advanced settings* section. Read more here.
  * backend: Added functionality to identify key frames using Gemini and extract them as additional Demand Gen image assets.
  * backend: Improved the extraction process to maintain consistency across the generated descriptions and keywords per segment.
* [December 2024]
  * frontend: Added functionality to generate Demand Gen text assets in a desired target language. Read more here.
  * frontend: Added possibility to load previously re

……（后略，见原文）