Templit is an agentic video production platform for movie trailers — think Lovable, but for movie trailers. An AI agent guides you through the entire video creation process in a conversational chat interface, from initial concept to a fully produced, stitched-together video.

HD Demo: https://youtu.be/8RdLiv3fTVU

Templit has two screens: a landing page where you enter your video idea, and the editor where the agent takes over.

The editor is a three-panel layout:

• Chat panel (left) — The AI agent lives here. It walks you through each step of the pipeline, asks for your input, and kicks off generation tasks behind the scenes.
• Preview panel (top right) — Watch your video with full playback controls once it's generated.
• Timeline panel (bottom right) — A multi-track timeline showing your video and audio clips, with a draggable playhead and zoom controls.

The agent walks you through six steps, each with an approval gate where you can review, request changes, or approve before moving on:

1. Project Overview — The agent collects your video's topic, duration, and aspect ratio.
2. Artistic Style — The agent suggests four tailored visual styles based on your concept. Pick one or describe your own.
3. Character Generation — Upload reference photos of your characters. The agent sends them to Google Nano Banana 2 which generates a four-view character sheet (front, back, left, right) for each one.
4. Script & Scenes — The agent proposes four storyboard concepts, then generates a full scene-by-scene script with timestamps and dialogue. Scene count is calculated from your chosen duration (~5 seconds per scene).
5. Scene Thumbnails — Two phases:
   • Scene Locations — Google Nano Banana 2 generates a background/environment image for each scene.
   • Scene Thumbnails with Characters — Google Nano Banana 2 composites your characters into each location to create the final scene thumbnails.
6. Final Video — Each scene thumbnail is sent to OpenAI Sora 2 (image-to-video) to generate a video clip. The clips are stitched together with FFmpeg into one continuous video and loaded into the preview player and timeline.

Throughout the entire pipeline, the agent maintains a Project Overview panel on the right side of the screen. This living document is automatically updated after every interaction — if you go back and change your artistic style, revise a scene, or edit dialogue, the agent silently updates the overview to reflect the current state of your project.

The agent is orchestrated by Gemini 2.5 Flash for all chat and reasoning, with the image and video generation models called through API routes as the pipeline progresses.

Prompt: Make a video to help me teach my kid how to make a grilled cheese sandwich.

Prompt: Make a video to help me teach my kid how to make a hamburger.

Prompt: Make a video about two friends learning how to make a grilled cheese sandwich.

• Next.js 16 with App Router and TypeScript
• Tailwind CSS v4 with a dark theme
• Zustand for state management
• Gemini 2.5 Flash — agent chat and reasoning
• Google Nano Banana 2 — character sheets, scene locations, and scene thumbnails (image generation)
• OpenAI Sora 2 — scene-to-video generation
• FFmpeg — video stitching
• react-resizable-panels for the editor layout

npm install
npm run dev

Then open http://localhost:3000.