Spica is a web application that extends OpenAI's Sora 2 video generation capabilities by using AI to plan multiple scene prompts with maximum continuity. Instead of generating a single short video, Spica intelligently breaks down your concept into multiple segments that seamlessly connect through frame-by-frame continuity.
Using GPT-4 (or similiar models) for intelligent prompt planning and Sora 2's input_reference feature, each segment begins exactly where the previous one ended, creating longer, more coherent video sequences that maintain consistent style, lighting, and subject identity throughout.
- AI-Planned Prompts - GPT-4 intelligently creates scene prompts with maximum continuity
- Frame-by-Frame Continuity - Each segment starts from the final frame of the previous one
- First Frame Reference - Upload a custom starting image for your first segment
- AMOLED Dark Theme - Beautiful true black interface with monospace typography
- Real-time Progress - Track generation progress for each segment
- Instant Download - Download your generated videos as soon as they're ready
- Fully Configurable - Control models, resolution, duration, and segment count
Visit spica.kuber.studio to use Spica instantly in your browser!
No installation required. Just enter your OpenAI API key and start creating.
- Node.js: 18+
- OpenAI API Key: With GPT-4 and Sora 2 access
- Clone the repository:
git clone https://github.com/Kuberwastaken/spica
cd spica- Install dependencies:
npm install- (Optional) Create a
.env.localfile:
cp .env.local.example .env.localEdit .env.local and add your OpenAI API key, or enter it in the UI.
- Run the development server:
npm run dev- Open http://localhost:3000 in your browser.
- Enter your OpenAI API key - Your key stays in your browser, never stored on servers
- Write your base prompt - Describe your video concept (e.g., "A car driving through a futuristic city")
- (Optional) Upload a first frame - Start from a specific image
- Configure generation settings:
- Planner Model: GPT-4o (recommended) or GPT-4o Mini
- Sora Model: Sora 2 or Sora 2 Pro
- Resolution: 1280x720, 1920x1080, or 720x1280 (portrait)
- Seconds per Segment: 4, 8, or 12 seconds
- Number of Segments: 1-20 segments
- Generate - Watch real-time progress for each segment
- Download - Get your final video instantly
Base Prompt → AI Planning → Segment Prompts → Sora Generation → Frame Extraction → Continuity Chain
-
Planning Phase
GPT-4 analyzes your base prompt and creates N scene prompts with detailed context for continuity -
Generation Phase
Each segment is generated with Sora 2, using the final frame of the previous segment asinput_reference -
Frame Extraction
The last frame of each video is extracted and used as the starting point for the next segment -
Continuity Chain
This creates a seamless flow where each segment naturally continues from the previous one
- Framework: Next.js 15 with App Router
- Language: TypeScript 5.7
- UI Components: Shadcn/UI
- Styling: Tailwind CSS with AMOLED theme
- Font: JetBrains Mono
- API Integration: OpenAI SDK (GPT-4 + Sora 2)
| Setting | Options | Notes |
|---|---|---|
| Planner Model | GPT-4o, GPT-4o Mini, GPT-4 Turbo | GPT-4o recommended for best results |
| Sora Model | Sora 2, Sora 2 Pro | Pro version for higher quality |
| Resolution | 1280x720, 1920x1080, 720x1280 | HD/Full HD landscape or portrait |
| Duration | 4s, 8s, 12s per segment | 8 seconds recommended |
| Segments | 1-20 | Total duration = segments × duration |
Cinematic Journey
A cinematic journey through a neon-lit cyberpunk city at night,
showcasing futuristic architecture and flying vehicles
Nature Transitions
A serene forest path transitioning through the four seasons,
from spring blossoms to winter snow
Product Showcase
Professional product showcase of a sleek smartphone, highlighting
its design features and elegant materials
Abstract Art
Abstract geometric shapes morphing and flowing in a minimalist
space with dramatic lighting
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.
Inspired by Sora Extend by Matt Shumer.
Built with Next.js, Shadcn/UI, and OpenAI's Sora 2.