Lucidify

Awaken your subconscious. Lucidify uses Gemini to curate your dreams and inner thoughts, turning ethereal fragments into clear, actionable insights.

Comment

Inspiration

"What if you could interpret your dreams, enter them, and control the narrative?"

I am a lucid dreamer. I have experienced moments where I could consciously control my dreams, and I have even had precognitive dreams that blurred the lines between reality and the subconscious.

This personal ability sparked a question: Could technology replicate this experience for everyone? Lucidify was born from my lifelong fascination with the subconscious. It is not just an app; it is a digital bridge designed to help users interpret their dream content and virtually "enter" and "steer" their dreams using the power of Generative AI.

What it does

Lucidify is an AI-powered subconscious curation tool.

  1. Dream Input: Users enter fragments of their dreams or vague thoughts.
  2. Hypnotic Scripting: Gemini analyzes the input and generates a personalized "Hypnotic Script."
  3. Auditory Immersion: Google Cloud TTS (Neural2-F) acts as your digital guide. It delivers the script with a soothing, paced cadence designed to induce a meditative state, syncing perfectly with the visual flow.
  4. Visual Manifestation: Using Veo, the system generates ethereal videos that visualize the user's subconscious state.
  5. Lucid Transition: The app guides the user through a "Lucid Transition," turning abstract visuals into clear, actionable "Subconscious Insights."

How we built it

We utilized an AI-native workflow (Antigravity IDE) to orchestrate the absolute latest from Google's AI and web ecosystem, building a futuristic experience with:

  • Frontend: Svelte 5, Vite 7, Tailwind CSS 4, Lucide Svelte.
  • Backend: SvelteKit (Serverless Functions) on Google Cloud Platform (GCP).
  • The AI Engine:
    • Google Gemini 2.5 Flash: For ultra-fast reasoning and script generation.
    • Google Veo 3.1 (via Google AI Studio): For high-fidelity dream video generation.
    • Imagen 4.0 Ultra: For thumbnail imagery.
    • Google TTS (Neural2-F): For the hypnotic voice guide.

Challenges we ran into (Technical Deep Dive)

We encountered significant engineering hurdles while orchestrating cutting-edge AI models on a serverless architecture. Here is how we solved them:

1. Real-time Video Generation vs. Serverless Limits

  • The Challenge: Generating high-fidelity video with Google Veo 3.1 takes time, often exceeding the 60s hard timeout of serverless platforms (Vercel), leading to wasted compute and generic errors.
  • The Solution: We implemented a Time-Bounded Polling Architecture.
  • Implementation: We switched to Vertex AI's predictLongRunning endpoint and built a backend polling loop monitored by a safety timer (55s limit). We utilize Server-Sent Events (SSE) to send PROGRESS heartbeats every 4 seconds, keeping the connection alive while the heavy rendering happens on Google's infrastructure.

2. Robust LRO Polling for Async API Models

  • The Challenge: Veo 3.1 operations are Long-Running Operations (LROs) with deeply nested, variable response structures. Standard polling often failed with 404 or 400 errors due to schema mismatches between preview and production models.
  • The Solution: A Recursive URI Extraction & Polling mechanism.
  • Implementation: Our system captures the operationName from the kickoff response and uses the v1beta endpoint to poll. We implemented a resilient JSON parser that recursively searches for video.uri across multiple potential fields (generatedSamples, response.video[0]), ensuring stability regardless of API schema shifts.

3. The "Swan Strategy" (Decoupled UX)

  • The Challenge: Complex backend polling (LROs, multiple API calls) risked making the frontend unstable and jittery.
  • The Solution: We decoupled the UX using the "Swan Analogy."
  • Implementation: Like a swan appearing graceful while paddling frantically underwater, our frontend displays a high-quality static state ("Warping Reality..."), while the backend handles the heavy lifting. We use SSE solely as a "heartbeat" to maintain the stream without triggering visual flickers, allowing us to migrate backend logic without breaking the UI.

4. High-Fidelity Cloud TTS & Key Scoping

  • The Challenge: Standard AI Studio keys don't support Google Cloud TTS, and the experimental Journey-F voices crashed with standard SSML parameters (pitch/rate).
  • The Solution: A Dedicated Multi-Key Architecture and Safe Parameter Protocol.
  • Implementation: We separated the architecture to use a dedicated GCP API Key for TTS services. We also migrated to the Neural2-F voice model, optimizing it with aggressive SSML pacing (rate="0.85", pitch="-4.0st") to achieve the deep, hypnotic tone required for the app.

5. Aesthetic Cohesion with "The Chroma Aura System"

  • The Challenge: Managing dozens of 3D-rotating, interactive cards with unique aesthetics in a performant way.
  • The Solution: The Chroma Aura System powered by Svelte 5 Runes.
  • Implementation: We created a deterministic gradient palette where each card's "soul" is derived from its index. Using Svelte 5's $state and $derived runes, we handle complex 3D CSS transitions on the GPU, ensuring a silky-smooth 60fps experience even with heavy visual elements.

Accomplishments that we're proud of

🌟 1. Architecting "The Dream Engine"

We built a robust multi-model pipeline that acts as a bridge to the subconscious.

  • Parallel Synthesis: We achieved perfect synchronization between audio and visuals. The backend triggers Voice Synthesis (TTS) and Prompt Refinement (Director) in parallel, ensuring the narrator speaks the exact script being visualized.
  • Smart Ambient Audio: We use Gemini 2.5 Flash to analyze the dream's emotional profile and intelligently select one of 5 ambient loops (Nature, Space, Horror, etc.), cross-fading them dynamically using a zero-clash sequence.
  • Aesthetic Sanitization: Gemini acts as a safety filter, automatically translating violent or sensitive content into artistic metaphors (e.g., "blood" → "rose petals") before manifestation.

🎥 2. Multi-Model Manifestation & Resilience

We didn't just rely on one model; we built a resilient fallback system.

  • Veo 3.1 Fast: The primary engine for cinematic dream shorts.
  • Imagen 4.0 Ultra Fallback: If video quotas are hit, the system gracefully downgrades to state-of-the-art static imagery.
  • Fixing the "Mist": We solved the "Only Mist" bug in Veo by implementing browser-side key appending and a theatrical transition sequence that clears the introductory fog.

🔮 3. The "Chroma Aura" Journal (Svelte 5)

We pushed the boundaries of the new Svelte 5 framework.

  • Runes Everywhere: The entire UI state is managed by Svelte 5 Runes ($state, $effect), creating a reactive and bug-free experience.
  • 3D Tactile Interaction: The Dream Journal features Tarot-style cards that levitate and flip in 3D space. Each card has a unique "Chroma Aura"—a dual-color gradient generated deterministically—giving every dream a unique visual soul.
  • Local Persistence: We implemented a robust localStorage system with a "Confirm Modal" (inspired by our other SaaS, Cubrain) to prevent accidental deletions of precious dream records.

🕹️ 4. Lucid Mode & The Easter Egg

  • Lucid Mode: We created an interactive "Take Control" flow. A "Lucid Transition" freezes time (grayscale + awakening.mp3), allowing the user to rewrite the narrative mid-dream.
  • Ghost Typing: We added a production-ready Easter Egg (F8 key) that simulates "Ghost Typing," auto-filling context-aware prompts for demo purposes.

🤖 5. The "Agentic" Development Process

We didn't just code; we orchestrated. Using Antigravity IDE, we leveraged Gemini to architect the backend logic while we focused on the creative vision, proving that a single maker can build enterprise-grade apps with AI assistance.

What we learned

We learned that Multimodal AI is not just for productivity; it's a powerful tool for self-reflection. We also mastered the art of asynchronous AI orchestration—making multiple models (Gemini, Veo, TTS) work together in harmony despite their different latency profiles.

What's next for Lucidify

We have a clear roadmap to evolve Lucidify from a web prototype into a daily habit-forming ecosystem.

1. The "Hypnopompic" Wake-Up Mode (Killer Feature)

We plan to solve the "forgotten dream" problem by integrating an alarm system.

  • The Flow: Stop alarm → Mic activates immediately.
  • Native Audio AI: Gemini 2.5 Flash captures groggy voice notes and instantly structures them into dream prompts.

2. The Cognitive Ecosystem: "The Subconscious Bridge"

Lucidify is the missing link in our cognitive suite.

  • Conscious Learning (Cubrain): Connects with Cubrain (our SaaS for active recall/flashcards).
  • Subconscious Insight (Lucidify): Cubrain structures awake knowledge; Lucidify structures asleep insights.

3. Mobile Expansion (TWA Strategy)

  • Phase 1 (Now): PWA implementation.
  • Phase 2 (Next): Packaging the SvelteKit app into an Android TWA (Trusted Web Activity) for Play Store launch, maintaining a single serverless backend.

4. Empathetic Interpretation & Therapy

Expanding the "Dream Architect" into a mental health tool to visualize anxieties in a non-threatening, metaphorical way.

Built With

Share this project:

Updates