IgniteAI - The First "Director-in-a-Box" for UGC Ads

IgniteAI builds complete, narratively driven campaigns with Multimodal Intelligence, Structured Storytelling & Real-Time "Director's View".

Comment

IgniteAI - The First "Director-in-a-Box" for UGC Ads

Inspiration

As developers and creators, we noticed a massive gap in the market. Small business owners know they need short-form video content (TikTok/Reels) to survive, but they are blocked by the "Creative Triangle of Death":

  • Cost: Agencies charge $500+ per video.
  • Skills: Learning Premiere Pro is hard and time-consuming.
  • Time: Ideation and filming take hours.

We wanted to build a "Director-in-a-Box" that doesn't just edit clips, but actually dreams up the entire campaign—from script to screen—using the latest multimodal AI.

What it does

IgniteAI is an end-to-end autonomous video ad generator.

  1. Ideation: You give it a simple prompt (e.g., "A high-energy coffee ad for Gen Z").
  2. Scripting: It uses Gemini 2.5 Flash to write a hooked-based script (Hook -> Feature -> CTA).
  3. Casting & Filming: It uses Gemini 3 Pro to generate high-fidelity distinct scenes and Google Veo to turn them into motion.
  4. Production: It assembles the video with music, transitions, and branding.
  5. Consumption: It features a "Community Reels Viewer" where you can scroll through generated ads just like on TikTok, to see what others are building.

How we built it

We built a modern, scalable full-stack application:

  • AI Core: We utilized Gemini 2.5 Flash for high-speed reasoning and scripting, and the new Gemini 3 Pro (via Vertex AI) for its superior visual understanding and image generation capabilities. Google Veo handles the video synthesis.
  • Frontend: Built with Angular 19, featuring a glassmorphic UI and a custom ReelsViewerComponent that uses the IntersectionObserver API for high-performance video autoplaying.
  • Backend: A Python FastAPI service deployed on Google Cloud Run. It handles the complex orchestration of AI agents.
  • Infrastructure: Firebase for Authentication and Hosting, GCP Secret Manager for security, and FFmpeg for high-speed video rendering.

Challenges we ran into

  • Consistency: Keeping the "actor" looking the same across different scenes is the holy grail of AI video. We solved this by implementing a "Visual DNA" system that passes character context (age, ethnicity, clothing) into every single prompt.
  • Performance: Generating 4-5 videos in parallel hit API rate limits. We implemented a robust queuing system and optimized IntersectionObserver on the frontend so we only load resources for the video currently on screen.
  • Legacy Code: Migrating from an older generation framework to the new Gemini 3/Veo pipeline required a significant refactor of our media_factory.py to support dynamic model selection.

Accomplishments that we're proud of

  • Gemini 3 Integration: We are one of the first apps to fully integrate Gemini 3 Pro for image generation, offering users a premium "High Fidelity" toggle that noticeably improves visual quality.
  • The "Reels" Experience: Building a custom video player that feels exactly like TikTok—smooth scrolling, snap-to-page, and instant playback—on the web was a tough frontend challenge that we nailed.
  • Real Utility: This isn't just a toy. We built a credit system, project history, and downloadable assets. It's a shipping product.

What we learned

  • Multimodal is Key: Text-to-Video is okay, but Image-to-Video (providing a product image) is game-changing. Gemini's ability to "see" the user's uploaded product and write a script about it is magical.
  • Latency Matters: Users hate waiting. We learned to stream logs via WebSockets so the user sees "Dreaming up visuals..." or "Composing music..." in real-time, making the 60-second generation time feel instant.

What's next for IgniteAI

  • Agency Mode: A dashboard for marketing agencies to manage multiple client brands.
  • Direct Publishing: Integration with TikTok/Instagram APIs to auto-post the generated videos.
  • Fine-Grained Control: A timeline editor to manually tweak the AI's cuts and transitions.

Built With

  • Angular 19
  • Python (FastAPI)
  • Google Cloud Run
  • Firebase
  • Gemini 2.5 Flash
  • Gemini 3 Pro
  • Google Veo
  • Vertex AI
  • FFmpeg

Built With

Share this project:

Updates

posted an update —

IgniteAI Update — Building the Director-in-a-Box

Over the past few days, IgniteAI’s video engine has been going through a major architectural upgrade.

We’re currently migrating the pipeline to a new V3 Parallel-Core architecture designed to generate UGC-style ad videos faster, more consistently, and with better creative control.

Here’s what’s new under the hood:

** Director → Scene Workers → Conductor architecture** Instead of generating everything sequentially, the system now behaves more like a real production crew:

  • A Director node decides the creative strategy
  • Parallel scene workers generate hook, feature, and CTA shots simultaneously
  • A Conductor stage assembles the final video with music, voice, and captions

** Parallel scene generation** Scenes are now rendered concurrently using a worker pool, significantly reducing total generation time.

** Strategy-aware ads** The engine now extracts:

  • target persona
  • pain point
  • hook angle

This allows IgniteAI to structure ads more like high-performing UGC creatives.

** Modular AI “skills” architecture** Video generation, captions, voice, images, and fallback animation are now handled by independent skills — making the system easier to extend and upgrade.

** Powered by Veo 3.1 + Gemini Image** The new pipeline fully leverages Google’s latest video and image generation models to create short-form ad content.

Still early days, but the goal is clear:

Turn IgniteAI into the first true “Director-in-a-Box” for UGC ads.

More updates soon as the V3 engine comes online.

Log in or sign up for Devpost to join the conversation.