Inspiration
Editing 30-sec ads in After Effects sucks. Prompt-based tools spit out junk or break. We wanted a deterministic, code-first way to get studio-quality spots from a single prompt.
What it does
Takes a prompt → analyzes industry, tone, style → picks scene templates → generates React/Remotion components → renders a 1080p MP4 locally or on AWS Lambda.
Enhanced with advanced capabilities:
- Smart file upload system - Drag & drop brand assets (logos, documents, images)
- AI color extraction - Real-time brand color analysis using k-means clustering
- Voice narration - Auto-narrates video content with browser TTS
- Brand consistency - Videos automatically match uploaded brand colors and assets
How we built it
- Frontend: Next.js 15 with elegant 50/50 layout, drag-and-drop file uploads, real-time color preview
- AI Engine: Claude 3.5 Sonnet as "director" (structure JSON) and "generator" (TSX code)
- Template System: 17+ advanced scene templates with dynamic color integration
- Color Analysis: Client-side k-means clustering for brand color extraction from uploaded images
- Voice Integration: Browser Speech Synthesis API for automatic video narration
- File Management: Robust upload system with image processing and document ingestion
- Rendering: Remotion with TransitionSeries, springs/interpolate, and particle systems
Challenges we ran into
- Getting Claude to always return valid JSON and compilable TSX
- Syncing scene timings to exactly 900 frames without drift
- Race conditions on parallel file writes and cleanup
- Color extraction performance - Processing large images without blocking UI
- File upload reliability - Handling multiple formats and error states gracefully
- Voice timing synchronization - Matching narration to video frame progression
Accomplishments that we're proud of
- End-to-end prompt → MP4 in a single command
- Advanced template system with 17+ reusable scenes (hero, features, stats, CTA, product showcases)
- Intelligent brand integration - Real color extraction and dynamic application
- Seamless file handling - Drag-and-drop with visual feedback and error recovery
- Voice narration system - Auto-reads video content with scene-by-scene timing
- Robust fallbacks and validation so renders rarely fail
- Smooth, modern motion that actually looks like an ad, not a slideshow
What we learned
- Advanced color theory - K-means clustering for dominant color extraction
- File processing patterns - Client-side image analysis without server overhead
- Strong prompting patterns for codegen and schema enforcement
- Deep Remotion internals: performance, transitions, particle systems
- Voice synthesis integration - Browser APIs for real-time narration
- How to structure a template system so designers and devs can both extend it
What's next for Admotion
- Enhanced brand kit processing - Font extraction, style guide analysis
- Professional voice integration - ElevenLabs/VAPI for studio-quality narration
- Advanced color harmonies - Complementary and triadic color scheme generation
- Variable lengths (15/30/60s) and adaptive timing
- Music/SFX library and VO alignment with uploaded audio tracks
- Live preview editor with timeline scrub + per-scene edits
- Real-time collaboration - Multi-user editing with brand asset sharing
- Caching, DB storage, and A/B variant generation for marketers
Built With
- 3.5
- amazon
- amazon-web-services
- api
- claude
- css
- framer
- lambda
- motion
- next.js
- node.js
- react
- remotion
- remotion/transitions
- s3
- sonnet
- tailwind
- typescript
Log in or sign up for Devpost to join the conversation.