• ## Inspiration
I built SoftSky.cloud because life gets noisy fast, and most people don't need more content, they just need one calm moment that feels personal. The idea was simple: instead of picking from a static library, what if someone could just describe the vibe they need right now and get a custom visual + sound space instantly?
## What it does
SoftSky.cloud lets a user type a short intention, pick a duration, and compose a full relaxing experience. It generates:
- a live visual scene (rendered in browser with PixiJS)
- matching ambient music (rendered in browser with Tone.js)
It also supports:
- playback controls (play/pause/restart/volume/fullscreen)
- theme + day/night styling
- shareable scene links
- Google sign-in for history and sharing
- generation history/library per signed-in user
## How I built it
I built it as a full-stack web app:
- Frontend: React + Vite
- Visual engine: PixiJS
- Audio engine: Tone.js
- Backend API: Node.js + Express on Google Cloud Run
- AI generation: Gemini (scene code + music code generation)
- Data/persistence: Firestore + Cloud Storage
- Async workflows: Cloud Tasks (share job processing)
- Auth: Google OAuth
The key architecture choice: I generate executable scene/music code, not heavy pre-rendered media files.
## Challenges I ran into Biggest challenges were very real production issues:
- Timeout + 504 issues during long generation paths
- Dealing with invalid generated code and strict validation loops
- Making playback smooth while scene rendering stayed stable
- Keeping everything secure while still moving fast
## Accomplishments that we're proud of
- Shipped a working end-to-end product quickly, from idea to live domain
- Real-time generated scene + music experience (not static preset playback)
- Built share, auth, history, and quota controls into a usable product
- Got both desktop and mobile experience feeling polished
- Put the app on scalable cloud infrastructure early instead of a demo-only setup
## What I learned
- Prompt quality and validation are everything in code-generation products
- Guardrails matter as much as creativity when generated code is executable
- Infra details (timeouts, DNS, CORS, auth callbacks) can make or break UX
- Responsive UI polish takes way more iteration than expected
- Parallel generation and clean runtime boundaries make a huge difference
## What's next for Skysoft.cloud
This is just the start. The bigger vision is a full virtual space creator powered by code generation:
- persistent personal spaces, not one-off sessions
- deeper personalization and saved “world states”
- inviting others into shared spaces
I strongly believe code generation in media is still underused, and this direction can stay highly personalized and cost-efficient compared to heavy pre-rendered pipelines.
Built With
- bash
- cloud-run
- cloud-run-domain-mapping
- cors
- css3
- eslint
- express.js
- gemini-3-flash-preview
- google-auth-library
- google-cloud
- google-cloud-tasks
- google-firestore
- google-genai-sdk
- google-oauth-2
- google-secret-manager
- html5
- javascript
- managed-tls
- namecheap-dns
- node.js
- npm
- pixi-js
- react
- tone-js
- vite
Log in or sign up for Devpost to join the conversation.