Me Time
AI-guided mindful creativity. Step by step. Breath by breath.
Powered by BRIA FIBO — JSON-native control meets agentic workflows for progressive visual guidance
🚀 Try the Live App • 🎥 Watch the Demo
💭 Inspiration
In our hyper-connected world, we've lost the art of slowing down. Meditation apps tell us to "just breathe," but many find empty stillness uncomfortable. What if mindfulness could be active?
Me Time brings together two powerful ideas:
- Mindful doing: Creative activities that quiet the mind through gentle focus
- Progressive guidance: AI that breaks down intimidating art into calming, achievable steps
We were inspired by art therapy's proven benefits for mental well-being but wanted to make it accessible to everyone—no expensive classes, no artistic background required. Just you, simple materials, and an AI guide that meets you where you are.
✨ What It Does
Me Time is an AI-powered creative companion that transforms any artistic idea into a meditative, step-by-step journey using BRIA FIBO's JSON-native control and progressive refinement.
Core Workflow
- Generate or Upload — Create a beautiful reference image with AI, or upload your own inspiration
- Refine & Explore — Use text prompts to iteratively refine any image until it feels right
- Break It Down — AI dissects the final image into calm, progressive steps
- Create Together — Follow visual guides that build naturally from simple to complete
How It Works: A Real Example
Imagine you want to learn watercolor painting but feel overwhelmed...
Step 1: Tell "Me Time" Your Idea
You: "I want to paint a peaceful watercolor sunset over a lake"
Step 2: AI Generates Your Master Reference
- The Watercolor Agent understands your medium-specific needs
- Creates a beautiful reference image using BRIA FIBO's JSON-native controls
- Captures the soft, transparent quality unique to watercolors
- Result: A stunning watercolor sunset appears on your infinite canvas ✨
Step 3: Refine Until It's Perfect (Optional)
You: "Make the clouds more dramatic"
You: "Add a small boat on the water"
- Each text prompt creates a new refined version
- FIBO's seed consistency ensures the same style/lighting
- Only the elements you mention change—everything else stays coherent
- Result: Your perfect vision, iteratively refined 🎨
Step 4: AI Breaks It Down Into Steps
- The Dissection Agent analyzes your final masterpiece
- Understands watercolor-specific techniques (wet-on-wet, layering, transparency)
- Generates 5 progressive steps showing how to build the painting naturally:
- Light wash - Initial sky gradient
- Sky colors - Orange and purple layers
- Cloud forms - Dramatic cloud shapes
- Water & horizon - Lake reflection and horizon line
- Final details - Boat and finishing touches
Step 5: Visual Step-by-Step Guide Created
- Each step image uses the same artistic seed (perfect visual consistency!)
- Every image looks like a natural progression of THE SAME painting
- You see exactly what your canvas should look like at each stage
- Result: A complete, calming visual guide from blank page to finished art 🌅
Your Canvas Now Shows:
- ✨ Your beautiful master reference
- 📋 5 step-by-step images showing natural progression
- 🔗 Visual connections showing the creative flow
- 💾 Ready to export as PDF or work from on-screen
All of this happens in ~30 seconds, fully automated by 13 specialized AI agents working together.
💡 Beyond Creative Guides: Me Time works with ANY image. Generate from scratch with FIBO AI,upload your own photos, or import images from other AI tools. While the example above shows instruction generation for calming activities, you can also use "Me Time" purely as a controlled step-by-step image editor connect text prompts to any image to refine specific elements (lighting, colors, objects) while preserving the overall composition. Think of it as Photoshop meets conversational AI, powered by FIBO's disentangled control.
10 Mindful Activities
| ✏️ Drawing | 🎨 Coloring Book | 🪵 Miniatures | 🧵 Fabric Art | 💐 Vase Design |
| 🌊 Watercolor | 🖼️ Oil Painting | 💎 Jewelry | 🔮 Patterns | 🎮 Characters |
Each activity has a specialized AI agent with calming, activity-specific guidance.
Key Features
- Infinite Canvas: Spatial workspace for organizing your creative journey
- Progressive Refinement: Images build naturally through FIBO's seed consistency
- Text-to-Image Refinement: Connect text prompts to any image for instant variations
- Upload & Refine: Start with your own photos and evolve them with AI
- Export Everything: Download complete visual guides as ZIP/PDF
🛠️ How We Built It
FIBO-Powered Multiagent Architecture
Me Time showcases FIBO's production-ready capabilities through a specialized agent system:
┌─────────────────────────────────────────────────────────────────────┐
│ 👤 USER INTERFACE LAYER │
├─────────────────────────────────────────────────────────────────────┤
│ ∞ Infinite Canvas (React Flow) 💬 Chat Interface │
│ • Image Nodes (Master + Steps) • Text Refinement Nodes │
│ • Visual Connections • Upload Interface │
│ • Export Controls (ZIP/PDF) • Activity Selector │
└────────────┬────────────────────────────────────┬──────────────────┘
│ │
▼ ▼
┌──────────────────────────────────┐ ┌─────────────────────────────┐
│ agentService.ts (Public API) │◀─│ CanvasWorkspace.tsx │
│ • generateMaster() │ │ • Node State Management │
│ • generateSteps() │ │ • Edge Connections │
│ • refineImage() │ │ • Text→Image Refinement │
└──────────────┬───────────────────┘ └─────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────┐
│ 🎭 AGENT ORCHESTRATION LAYER │
├─────────────────────────────────────────────────────────────────────┤
│ AgentOrchestrator.ts │
│ Routes by activity type → Specialized Agent │
├─────────────────────────────────────────────────────────────────────┤
│ ✏️ Drawing 🎨 Coloring 🌊 Watercolor 🖼️ Oil Painting │
│ 🪵 Miniature 🧵 Fabric 💐 Vase 💎 Jewelry │
│ 🔮 Patterns 🎮 Characters │
├─────────────────────────────────────────────────────────────────────┤
│ CategoryAgentBase.ts (Shared Logic) │
│ • getMasterImagePrompt() - Activity-specific master prompts │
│ • generateStepImage() - Progressive refinement logic │
│ • createRefinementInstruction()- Context-aware step prompts │
└──────────────┬──────────────────────────────────┬─────────────────┘
│ │
▼ ▼
┌──────────────────────────────┐ ┌────────────────────────────────┐
│ promptEngineering.ts │ │ briaService.ts │
│ (Gemini 2.5 Flash VLM) │ │ (BRIA FIBO API) │
├──────────────────────────────┤ ├────────────────────────────────┤
│ • Prompt → Structured JSON │───▶│ • generateImage() │
│ • Activity-aware reasoning │ │ ├─ VLM-to-JSON translation │
│ • 1000+ word prompts │ │ ├─ Returns structured JSON │
│ • Safety & aesthetics │ │ └─ Returns seed │
│ │ │ │
│ • dissectIntoSteps() │ │ • refineImage() │
│ ├─ Analyze final image │ │ ├─ Same seed for coherence │
│ ├─ Generate 5 steps │ │ ├─ Modify JSON controls │
│ └─ Materials & safety │ │ └─ Progressive refinement │
│ │ │ │
│ • refineStructuredPrompt() │ │ • generateStructuredPrompt() │
│ └─ Text → JSON mods │ │ └─ Extract from uploads │
└──────────────────────────────┘ └────────────────────────────────┘
│
▼
┌──────────────────────────────────────┐
│ BRIA FIBO External API │
├──────────────────────────────────────┤
│ • JSON-native image generation │
│ • Seed-based consistency │
│ • Disentangled refinement │
│ • VLM structured prompt extraction │
└──────────────────────────────────────┘
DATA FLOW EXAMPLES:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1️⃣ MASTER GENERATION (from scratch)
User Prompt → Orchestrator → WatercolorAgent
→ VLM (prompt engineering) → Structured JSON
→ FIBO API → {imageUrl, structuredPrompt, seed}
→ Canvas (Master Node)
2️⃣ TEXT REFINEMENT (existing image)
Text Node + Image Node → agentService.refineImage()
→ VLM (refine structured prompt) → Modified JSON
→ FIBO API (same seed + new JSON) → Refined Image
→ Canvas (New Image Node connected via edge)
3️⃣ STEP DISSECTION (progressive build)
Master Image → CategoryAgent.dissectIntoSteps()
→ VLM (analyze + create 5 steps)
→ For each step: CategoryAgent.generateStepImage()
→ VLM (calculate % complete) → Modified JSON
→ FIBO API (same seed + step JSON) → Step Image
→ Canvas (5 Step Nodes connected to Master)
4️⃣ UPLOAD + REFINE (user image)
Upload File → FIBO.generateStructuredPrompt()
→ Extract JSON from image → Canvas Node
→ Can now refine via Text Nodes (flow #2)
Technical Pipeline
1. Image Generation (AI or Upload)
// Option A: Generate from prompt
const structuredPrompt = await VLM.generateJSON(userPrompt, activity);
const result = await FIBO.generate(structuredPrompt);
// Returns: { imageUrl, structuredPrompt, seed }
// Option B: Upload existing image
const uploadedImage = await uploadToCanvas(file);
// Use FIBO's structured prompt extraction for refinement
2. Progressive Refinement (The Innovation)
// Text-based refinement using FIBO's seed + structured prompt
const refinedResult = await FIBO.refine(
originalStructuredPrompt, // Preserve base composition
originalSeed, // Ensure visual consistency
"add cherry blossoms" // What to change
);
// Same background, lighting, style — only requested change applied
3. Step Dissection
// VLM analyzes final image, generates calming instructions
const steps = await VLM.dissect(finalImage, activity);
// Returns: materials, 5-step progression, safety tips
// Generate step images using FIBO refinement
for (let step of steps) {
const stepImage = await FIBO.refine(
masterStructuredPrompt,
masterSeed,
createGentleInstruction(step, completionPercent)
);
}
Activity-Specific Agents
Each agent provides calming, contextual guidance:
| Activity | Agent Approach |
|---|---|
| Watercolor | "gentle washes build transparent layers" |
| Drawing | "light sketching lines find their form" |
| Coloring | "colors flow into outlined spaces" |
| Patterns | "repeating motifs emerge naturally" |
This isn't generic AI—it's mindful, activity-aware guidance.
🏔️ Challenges We Ran Into
- Visual Consistency Across Steps — Ensuring each step image felt like the same artwork required mastering FIBO's seed + structured prompt workflow
- Upload Integration — Enabling users to refine uploaded images (not just AI-generated) needed careful handling of structured prompt extraction
- Meditative Pacing — Balancing AI capability with simplicity; we removed complexity to preserve the calm experience
- Text-to-Image Connections — Building an intuitive canvas UI where text prompts visually connect to images they refine
🏆 Accomplishments That We're Proud Of
- Production-Ready Agentic Workflow: 10 specialized agents orchestrated seamlessly
- Hybrid AI + Human Input: Equally powerful whether generating from scratch or refining uploads
- Progressive Refinement Mastery: FIBO's seed consistency creates visually coherent step sequences
- Calm-First UX: Every interaction designed for mindfulness, not speed
- Infinite Canvas Innovation: Spatial interface that mirrors the creative thinking process
📚 What We Learned
- FIBO's VLM-to-JSON translator handles 1000+ word prompts beautifully—this enabled rich, nuanced image control
- Seed + structured prompt is the key to consistency; same base + refinement instruction = perfect progressive builds
- Agent specialization matters: Generic prompts can't capture the nuance of watercolor vs. oil painting
- Calm requires intentionality: We removed features that created anxiety (timers, complex menus) to preserve meditative flow
🌱 What's Next for Me Time
- Guided Sessions: 10-minute creativity sessions with breathing cues and ambient music
- Community Library: Share and discover calming projects from other creators
- HDR Support: FIBO's 16-bit color space for professional watercolor/painting workflows
- Voice Guidance: Optional narrated instructions for eyes-free creation
- Progress Journaling: Visual timeline of creative growth over time
Log in or sign up for Devpost to join the conversation.