Inspiration
Launching a physical innovation—whether it is an artisanal food-tech robot, a specialty coffee micro-roaster, or an autonomous cinema rig—usually requires hiring an expensive Hollywood ad agency.
Pre-production alone takes weeks: conducting market research, writing scripts, rendering storyboards, budgeting shoot logistics, and casting actors. For independent creators and small businesses, this $50,000+ barrier prevents groundbreaking ideas from reaching global audiences.
We asked: **What if an autonomous multi-agent production studio could turn any physical innovation into a grounded, broadcast-quality commercial production in 60 seconds?**
Inspired by our Salvadoran cultural heritage and the dream of automating traditional culinary art, we chose **La Pupusera 3000** (an automated artisanal pupusa maker) as our flagship launch
vehicle—combining authentic centuries-old tradition with cutting-edge Google Cloud AI.
---
## What It Does
**CineAd Studio** is an autonomous commercial director and production house that orchestrates specialized AI agents across the entire commercial lifecycle:
1. **Market & Cultural Grounding (Scout Agent):** Connects to the **Parallel Search API** to execute real-time web research, extracting consumer pain points, target demographics, and authentic
cultural anchors (e.g. Salvadoran quesillo melt, comal temperature curves).
2. Cinematic Screenwriting (Director Agent): Uses Gemini 2.5 Flash on Vertex AI to craft a professional 3-act narrative screenplay with camera movement, visual descriptions, and dialogue.
3. Visual Storyboarding (Visualizer Agent): Directs 16:9 cinematic storyboard panels with detailed camera lenses and lighting directives.
4. Technical Package & Budgeting (Producer Agent): Compiles an authentic camera package (ARRI Alexa, anamorphic lenses), lighting equipment, a crew call sheet, and a line-item budget
breakdown ($12,400 USD).
5. Broadcast Video Engine (Video Producer Agent): Synthesizes neural voiceovers via Google Cloud Text-to-Speech (es-US-Journey-F) and edits them using FFmpeg with camera push-in
motion (zoompan) and ambient score mixing into an exportable .mp4 commercial.
6. Interactive Multimodal Casting Booth (Casting Director Agent): Users can step onto the screen! Using multimodal Gemini 2.5 Flash, the agent inspects their facial demeanor and vocal
vibe, invents a creative commercial alter-ego (e.g. El Ciber-Comalero 2077), scores their screen charisma, and lets them star as the lead spokesperson in a personalized commercial spot.
---
## How We Built It
* **Google Cloud Vertex AI (`gemini-2.5-flash`):** Powers the core reasoning, structured JSON extraction, screenplay drafting, call-sheet scheduling, and multimodal facial/audio analysis.
* **Parallel Search API (`parallel-web`):** Executes live web research to discover real-time consumer data and cultural context for the Parallel Partner Track.
* **Google Cloud Text-to-Speech:** Synthesizes natural, broadcast-grade neural Spanish narration with Journey-grade vocal models.
* **Google Cloud Storage (`gs://cinead-studio-394455-assets`):** Hosts all compiled video deliverables, animatics, and storyboard assets.
* **Google Cloud Run:** Serverless deployment scaled to zero (`--min-instances=0`), ensuring zero idle cost and cloud-native reliability.
* **FFmpeg 7.1 & Python:** Automated multi-track audio ducking, kinetic camera motion (`zoompan`), letterboxing, and video concatenation.
* **Streamlit Front-End:** A dark cinema-themed studio interface featuring live pipeline telemetry, a Studio Campaign Vault, and an interactive casting booth.
---
## Challenges We Ran Into
* **Ensuring Strict Cultural Grounding:** Generic LLMs often produce stereotypical commercial scripts. Integrating the **Parallel Search API** solved this by feeding real market data and
authentic Salvadoran culinary terminology directly into the director's prompt context.
* Multimodal Face & Voice Synchronization: Combining user webcam snapshots with synthesized neural audio and dynamic FFmpeg camera motion required fine-tuning audio delay buffers (adelay)
and volume ducking so voiceovers remained crystal clear over ambient soundtracks.
* Cost & Serverless Constraints: We configured strict Google Cloud budget alerts ($10 cap) and scale-to-zero Cloud Run instances so the entire multi-agent commercial pipeline costs less than
$0.002 per run.
---
## Accomplishments That We're Proud Of
* **True Multi-Agent Coordination:** Creating a deterministic 5-agent pipeline that produces everything from research to camera packages, budgets, and video in under a minute.
* **Cultural Representation:** Proving that AI cinema can celebrate and preserve indigenous culinary heritage (*La Pupusera 3000*) rather than just generic corporate products.
* **Interactive Casting Experience:** Bringing creators directly into their own commercials through multimodal face analysis and creative persona generation.
---
## What We Learned
* Multi-agent chains with structured JSON outputs drastically outperform monolithic prompts in film production workflows.
* Real-time search grounding (Parallel Search API) is critical for commercial credibility.
* Google Cloud Vertex AI and Cloud Run provide the perfect enterprise infrastructure for high-throughput media generation.
---
## What's Next for CineAd Studio
* Direct integration with Google Cloud Veo for neural text-to-video generation once generally available.
* Automated multi-language localization (generating English, Spanish, and Mandarin cuts simultaneously).
* Direct publishing integration with YouTube Shorts and TikTok Ads Manager.
Built With
- gemini
- gemini2.5
- google-cloud
- vertex
Log in or sign up for Devpost to join the conversation.