Beta University AI Lab: Seed Agents Challenge

One topic in. One narrated short film out.

StoryCast is an autonomous multimodal agent that turns a plain-English topic into a finished one-minute video. Seed 2.0 plans the story, Seedream 5.0 draws the frames, Seedance 2.0 animates them, TTS narrates them, and ffmpeg assembles the final cut.

Why this is competitive

Visible agentic execution

The pipeline is not a black box. Judges can see scene planning, keyframe generation, narration, animation, and final assembly as separate steps with concrete artifacts.

Strong alignment with the public rubric

StoryCast is built around the current published weighting: Video Output Quality, Agentic Execution, and Demo & Presentation.

GitHub-native proof

The project is public, CI is green, and a complete render was executed inside GitHub Actions. That makes the prototype easy to inspect and easy to trust.

The current public challenge page on April 18, 2026 lists four challenge areas. StoryCast fits best under Track 01: AI Video Agents.
60s

Target final film duration

2m

Target submission demo length

1

Successful GitHub render artifact attached to the public repo workflow run

Pipeline

Seed 2.0 Seedream 5.0 Seedance 2.0 TTS + ffmpeg
  • Topic to structured scene blueprint
  • Blueprint to storyboard keyframes
  • Keyframes to image-to-video clips
  • Narration to voiced scenes
  • Scene clips to finished film

Suggested 2-minute structure

0:00-0:20

Hook

Open with the final rendered video first. Make the outcome obvious before you explain the system.

0:20-1:05

Live flow

Show the prompt entry, the agent stepping through scene planning and generation, and the resulting manifest or progress state.

1:05-1:35

Architecture

Use the repo diagram to explain why each model exists and how the pipeline stays modular and debuggable.

1:35-2:00

Vision

Position StoryCast as an autonomous explainer engine for education, science media, and branded storytelling.