What it does

ShortyRoll turns keywords into engaging short videos with one click.

Inspiration

Short-form video is painful to make from scratch. Existing AI tools take minutes (at best) to render a basic clip, which kills fast iteration. We wanted a one-tap path from keyword to watchable video in seconds, so creators can try, tweak direction, and try again without waiting.

How we built it

Built on IMG.LY’s Creative Editor SDK, ShortyRoll uses a fast, progressive stream pipeline so your video starts playing in seconds while the rest finalizes in the background.

Challenges we ran into

Our biggest constraint is operating cost. High quality TTS dominates unit economics at scale. To keep a free tier viable without slowing the experience, we default to on-device TTS for free usage while reserving richer cloud voices for Premium, all while maintaining smooth timing and caption sync during progressive playback.

Accomplishments that we're proud of

Our pipeline generates videos up to 6x real-time on the fastest models, with Premium at least 2x real-time, so playback starts immediately and runs uninterrupted while the rest finalizes in the background.

What we learned

We started with a classic flow. Generate the entire video behind a loading screen, then play it. It worked, but the wait killed momentum. While profiling the pipeline, we tried on-device TTS as a fallback and realised it was so fast that narration could keep up with scene assembly. That flipped our approach and instead of hiding progress, we stream the video as it comes together, with captions first and voice joining almost immediatly after.

What's next for ShortyRoll

  • We’re adding more video types (countdowns, comparisons, tips lists, explainers, before/after, myth-vs-fact, timelines)
  • letting you drop in your own images for personalized stories, and rolling out image-to-Quiz (add a photo, get Q&A rounds with reveals)
  • We’ll keep expanding voices and styles while staying instant

Built With

Share this project:

Updates