• 

  interface

Video Editor Agent - Project Story

Inspiration

• Cawd Bot (OpenClaw) and Remotion showed what was possible.
• Real editing pain: slow manual cuts, repetitive motion-graphics work, and too many tool switches.

What it does

• Edits videos using a Gemini-powered agent.
• Accepts a natural-language request, plans the edit, writes Remotion code, and renders outputs.

How we built it

• Gemini 3 for coding for the core agent reasoning and coding. It drives the workflow end-to-end:
  • Plans editing steps and timelines.
  • Generates Remotion React components.
  • Fixes issues and retries renders when needed.
• Multimodal Gemini capabilities for understanding video, image, and audio inputs:
  • Video analysis for scene understanding and key moments.
  • Image inspection for overlays, graphics, and style cues.
  • Audio inspection for timing and pacing.
• LangChain + LangGraph for orchestration, tool calling, and structured agent loops.
• Remotion as the video engine, producing consistent, programmable edits.
• Google Cloud VM to run the service with stable compute and fast deployment.

Challenges we ran into

• Rendering can be slow depending on complexity, assets, and machine load.
• Occasional model load hiccups (Gemini availability or cold-start latency).

Accomplishments that we're proud of

• A clean, usable web interface that makes editing approachable.
• The agent can generate and render complete videos from a single prompt.

What we learned

• Gemini's multimodal understanding is strong for video, image, and audio context.
• Prompt and context engineering matters a lot for consistent editing results.

What's next for Video Editor Agent

• Expand tooling and skills for more editing styles and effects.
• Improve reliability and speed for rendering.
• Collaborate directly with video editors to refine the workflow.

Built With

• fastapi
• gemini
• google-cloud
• langchain
• node.js
• python