Inspiration

EvoBridge explores a small creator's workflow: turn a brief into a visible storyboard, refine individual scenes, and export a real video while keeping control of the edits.

What it does

The studio asks for a theme, audience, duration and tone, then creates a three-scene storyboard. It supports an explicit English/Chinese command grammar such as "Shorten scene 2", "Tone warm" and "Undo", alongside direct scene controls. Before rendering, the creator confirms the exact revision and duration. Pillow and FFmpeg produce a real silent H.264 MP4 in landscape or portrait. Project history, cancellation, explicit retry, duplicate-request handling and SHA-256 manifests refer to stored jobs and output files.

Alexa+ simulation disclosure

This is an Alexa+ experience simulation built with AI-assisted development tools. The interaction layer uses bounded deterministic rules. It does not connect to the Alexa+ service, a device, an Alexa SDK or a runtime language model. The video output is motion graphics, not neural text-to-video. Unsupported commands produce an explanatory error. The live CSS preview is illustrative; the rendered-video view plays the actual MP4.

How it was built and what changed

EvoVid existed before the hackathon. This independent competition edition was built on October 11 from its public MIT branch, using the documented upstream snapshot 31255c83ff1f57143c5cf819c6e7a6bf8fc12fbe. The local SQLite job store and renderer concepts were retained or adapted. The bilingual web studio, multi-turn brief collection, revision-based conversational edits, explicit confirmation and export history are the substantive additions. The application uses Python, SQLite, Pillow, FFmpeg and browser HTML/CSS/JavaScript. OpenAI AI coding assistance supported development, documentation and testing. No private master-project data, credentials, personal photographs or model weights are included.

Demonstration and validation

The 2:33 video is an edited, sampled walkthrough of real browser captures with English captions. It shows a brief becoming a 12-second storyboard; scene, caption, tone and aspect-ratio edits; undo; an unsupported command preserving the saved revision; confirmation; and an actual 11-second video rendering and playing. The complete rendered output is included at normal speed. Persistence and restore-as-new are demonstrated. The application ran in a temporary private Linux GitHub Codespace. The hosted preview needed a one-line, uncommitted exact-origin compatibility change; public source was unchanged. This is not evidence of execution on a personal Windows computer or an Alexa device. The current source passed 17 unit tests. Real browser checks covered editing, undo, confirmation cancellation, rendering, playback, download/hash integrity, persistence, restore, repeat-render reuse, running-job cancellation and explicit retry. Earlier HTTP checks and DOM/API simulations are separate evidence. Native Windows/macOS launch, physical mobile-device layout and personal-image upload remain unverified.

Challenges and lessons

Revision snapshots prevent a render from silently using later edits. Cancellation must stop before an incomplete movie is published. Chinese captions require an appropriate installed font; font files are not redistributed. A stale unsupported-command toast currently remains until page reload and can overlap render progress. The demonstration preserves and labels this issue. The bounded interaction makes both working behavior and limitations inspectable.

Source and scope

The public MIT-licensed source and local setup instructions are available at https://github.com/2929496964-wq/evovid-contest-editions/tree/main/01-amazon-evobridge. This entry uses the Alexa+ simulated-experience route. It does not claim real Alexa integration or enter the optional AWS Builder or Open Source mini-challenges.