HACKATHON DEMO
AI Product
Ad Generator
Turn a product brief into a multi-scene video ad with planning, generation, evaluation, editing, and cloud history.
Built with Next.js · Gemini · Veo · Google Cloud · FFmpeg · ElevenLabs · Lyria
Shabad Vaswani
What the product does
A guided studio for creating product ads from a single brief.
Input
User enters a product brief, picks a mode, and can upload a product reference image.
Generate
Agents plan the storyboard, create structured prompts, generate clips, and evaluate attempts.
Control
User reviews scores, edits JSON prompts, selects candidates, trims clips, and rebuilds the final ad.
The goal
Make product-ad creation simple while keeping every scene visible, editable, and recoverable.
Instead of a black-box generation button, the app shows the plan, prompts, attempts, ratings, and final assembly process.
AI Product Ad Generator · Hackathon Submission · 2
User journey
The workflow is designed for fast iteration after generation.
1
Brief + reference
Describe product and ad goal; upload optional product image.
2
Storyboard
AI creates problem, reveal, demo, and hero scenes.
3
Structured prompts
Scene JSON captures subject, action, camera, lighting, negatives.
4
Candidates
Generate clips, keep every attempt, and show quality ratings.
5
Final draft
Merge selected clips with voice-over and music.
User stays in control
Edit structured JSON
Regenerate a scene
Select any candidate
Trim a stronger segment
Remove or restore clips
AI Product Ad Generator · Hackathon Submission · 3
System pipeline
Agents specialize in planning, prompt control, quality checks, and final assembly.
AI Director
Understands brief
and plans scenes
Story Compression
Keeps the ad
simple and tight
Clip Master
Compiles JSON into
Veo-safe prompts
Veo
Generates video
candidates
Evaluator + Repairer
Scores clips and
repairs prompts
FFmpeg + Audio
Merges clips with
VO and music
Key idea: each stage produces visible metadata, so the UI can show what happened and why.
AI Product Ad Generator · Hackathon Submission · 4
Studio controls
The frontend exposes the generation process instead of hiding it.
What the user can inspect
Structured visual prompt JSON
Compiled Veo prompt and negatives
Generated attempts and ratings
Evaluator issues and repair changes
Draft inclusion and final preview status
AI Product Ad Generator · Hackathon Submission · 5
Reliability layer
The product keeps the user moving even when a scene needs work.
Best-effort drafts
Every generated clip can be included in a draft with honest warnings and ratings.
Temporal salvage
Evaluator can suggest a stronger segment from inside a clip, then create a trimmed candidate.
Full history
Rejected clips, prompts, evaluations, and repairs remain inspectable and selectable.
Quality does not hide the output. It explains it.
The evaluator ranks attempts. The user decides what stays in the final ad.
100
attempts preserved
0→100
normalized score
3
auto retries per scene
∞
user regenerations
AI Product Ad Generator · Hackathon Submission · 6
Technical stack
A full-stack workflow for generation, editing, storage, and playback.
Next.js
Studio UI + API routes
Gemini
Director, evaluator, repairer
Veo
Video generation candidates
Google Cloud Storage
Media + run metadata
FFmpeg
Trim, concat, mux, fallback drafts
ElevenLabs + Lyria
Voice-over and music bed
Cloud history sidebar reloads previous runs from stored metadata, not in-memory state.
AI Product Ad Generator · Hackathon Submission · 7
What is next
The roadmap focuses on making final ads more polished and more controllable.
1
9:16 Instagram output
Add real vertical generation or safe crop/export path.
2
Image-generated hero cards
Use image models for accurate final product packshots.
3
Temporal trim editor
Let users keep the strongest segment of a generated clip.
Submission summary
A controllable AI ad studio: the app plans the ad, generates clips, evaluates quality, preserves every attempt, and lets the user edit or rebuild the final video.
Thank you
AI Product Ad Generator · Hackathon Submission · 8