Inspiration
I'm an actor and a TikTok creator from Israel with over 75,000 followers. Over the years I've edited hundreds of short-form videos and spent countless hours trying to understand why some moments make people stop scrolling while others don't.
Most editing tools focus on transitions, effects, and templates. But from my experience, that's not what makes a video viral.
The story comes first.
That became the idea behind ViralEngine: instead of asking AI to edit a video directly, I wanted to build an AI Director that understands the story first and lets a deterministic editing engine decide how every moment should be edited.
What it does
ViralEngine turns a raw conversation into a complete TikTok-ready short.
It first transcribes the video, understands the story, finds the hook, identifies the most important moments, and then builds a complete editing plan.
Instead of randomly applying effects, every editing decision is based on the importance of the story and the visual context of the original video.
The result is a vertical short that feels edited by a human rather than generated automatically.
How I built it
I intentionally separated the system into two independent stages.
The first stage is an AI Director whose only job is understanding the story:
finding the hook identifying the premise selecting the narrative ranking semantic importance
After that, everything becomes deterministic.
The local engine analyzes production information such as:
faces speaker changes pauses existing music transitions zooms reaction quality
For every candidate moment, the engine calculates multiple scores including story importance, reaction potential, surprise, energy and category bonuses.
These values are combined into a PeakScore that identifies the real highlights of the story.
Every editing package competes directly against no_effect, because sometimes the best editing decision is doing nothing.
Only after all of these decisions are made does the renderer create the final video.
Challenges I ran into
The hardest problem wasn't rendering videos.
It was deciding which moments actually deserve attention.
Real conversations are messy.
People interrupt each other.
Faces disappear.
Sometimes both speakers appear together.
Sometimes the original video already contains zooms, music or transitions.
The editor had to understand all of these situations before adding new effects while still producing deterministic and explainable results.
Accomplishments that we're proud of
I'm especially proud of separating semantic understanding from deterministic editing.
Instead of asking AI to generate a finished video, ViralEngine first understands the story and then applies explainable editing decisions.
I'm also proud of the impact scoring system, the explainable decision engine, and the production-aware editing pipeline that avoids conflicting with effects already present in the source video.
What I learned
This project taught me that good editing isn't about adding more effects.
It's about understanding why a moment matters.
Once the story is understood correctly, editing becomes a series of explainable decisions instead of random visual effects.
That idea became the foundation of ViralEngine.
What's next for ViralEngine
The current version focuses on conversational videos.
Next I want to support more video formats, generate captions automatically, improve multilingual understanding, and continue training the editing policies using feedback from real creators.
My long-term goal is to build an AI editor that understands storytelling well enough to become a true creative assistant instead of just another automatic video editor.
Log in or sign up for Devpost to join the conversation.