Ai Video Ad generation

Inspiration

Creating a short product ad usually requires several tools for writing, video generation, editing, voice-over, music, and review. We wanted to bring these steps into one workflow where AI handles most of the process while the user still controls every scene.

What it does

AI Video Ad Generator turns a product brief and optional reference image into a complete multi-scene advertisement.

It creates a storyboard, generates structured prompts, produces video clips, evaluates different attempts, and selects the best available clips. Users can review ratings, edit prompts, regenerate, trim, select, remove, or restore clips before rebuilding the final ad.

The selected clips are combined with voice-over and background music into one downloadable video.

How we built it

We built the application with Next.js and Google Cloud.

Gemini acts as the Ad Director, Story Compression agent, Clip Master, evaluator, and prompt repair agent. Each scene is represented as structured JSON containing the subject, action, environment, camera, lighting, style, and negative instructions.

Veo generates the video clips. FFmpeg trims and combines them, ElevenLabs creates voice-over, and Lyria generates the music. Run metadata, prompts, candidates, ratings, and videos are stored in Google Cloud Storage.

Challenges we ran into

The main challenge was coordinating several AI and media-processing stages in one reliable workflow.

We needed to manage structured prompts, multiple video attempts, evaluation scores, prompt repairs, product references, clip selection, audio timing, cloud storage, and final video assembly.

We also needed the application to preserve every generated candidate so users could review and control the result instead of receiving only one final output.

Accomplishments that we're proud of

We created an end-to-end system that goes beyond generating a single video clip.

The platform can plan an advertisement, generate multiple scenes, evaluate each attempt, repair prompts, preserve rejected clips, and build a complete draft even when some clips receive lower ratings.

We are also proud of the structured prompt editor, candidate comparison interface, persistent run history, video trimming workflow, and the ability to update only one scene without restarting the entire project.

What we learned

We learned that building an AI video product requires more than connecting a prompt to a video model.

The system needs clear scene planning, structured data, evaluation, retries, media processing, cloud persistence, and user controls. We also learned that AI agents work better when each agent has one focused responsibility and communicates through validated structured outputs.

What’s next for AI Video Ad Generation

Next, we want to improve vertical 9:16 video support, duration-aware storytelling, and image-generated product hero frames.

We also plan to improve automatic clip trimming, add smoother transitions, support more editing controls, and make product appearance more consistent across scenes.

Our longer-term goal is to let users create polished and customizable product ads from a simple brief without needing professional video-editing experience.

Built With

• gemini
• veo