ReelCraft — AI Showrunner

ReelCraft turns a few plain words into a fully directed short film, automatically enriching your prompt into cinematic direction, then generating and assembling the video with Qwen and Wan AI.

Comment

Inspiration

Most AI video tools take your words literally. Type "a girl walking down the aisle" into a raw video model and you get a flat, generic clip. But a real filmmaker would see a story there: the light, the camera movement, the emotion. I wanted to build the director that ordinary users don't have.

What it does

ReelCraft is an AI Showrunner. You type a few plain words. Behind the scenes, it runs a four-stage autonomous pipeline:

  1. Enrichment — rewrites your bare prompt into detailed cinematic direction, detecting the vibe (tender, funny, tense, action) and applying the matching camera, lighting, and color language.
  2. Script — writes a short visual script in that vibe.
  3. Storyboard — locks a consistent character and world, then breaks the story into single-character shots for visual consistency.
  4. Video + Assembly — generates each shot with Wan and stitches them into one finished film.

How we built it

  • Backend: Python + Flask, deployed on Alibaba Cloud ECS
  • AI: Qwen (qwen-plus) for enrichment, scripting, and storyboarding; Wan (wan2.2-t2v-plus) for video — all through Alibaba Cloud DashScope
  • Frontend: React, served via Nginx on the same Alibaba Cloud server
  • Assembly: MoviePy + FFmpeg The core differentiator is the enrichment engine, it's what separates ReelCraft from calling a raw video model.

Challenges we ran into

This was my first time deploying a full stack to cloud infrastructure. Getting Alibaba Cloud ECS provisioned, configuring Nginx, opening the right ports, and running Flask as a persistent systemd service was a steep but rewarding learning curve. Video assembly is memory-intensive, which pushed the limits of the free-tier instance.

Honestly, it was extremely disappointing; I almost gave up several times.

Accomplishments that we're proud of

  • Built a genuine differentiator: an enrichment engine that transforms a plain 6-word prompt into professional cinematic direction, automatically detecting vibe and applying matching camera, lighting, and color language.
  • Solved character consistency across independently-generated video shots using a locked "story bible" reused across every shot.
  • Took the project all the way to a live, publicly accessible deployment on Alibaba Cloud, frontend and backend both running on real cloud infrastructure
  • Designed the full four-stage autonomous pipeline (enrich → script → storyboard → video → assembly) end to end.

What we learned

How to take an idea from a local script all the way to a live, deployed, publicly accessible product on real cloud infrastructure and how prompt engineering, applied with a creator's eye, can dramatically change AI output quality

What's next for ReelCraft — AI Showrunner

Parallel video generation for speed, voiceover narration using Qwen's audio models, and true multi-part story continuation.

Built With

Share this project:

Updates