Ultramodern control center UI for orchestrating full-cycle LoRA training tailored to Flux1.dev carton generators and Wan2.2 cinematic diffusion. Inspired by ostris/ai-toolkit but redesigned as a neon cockpit for dataset ingest, auto-captioning, experiment management, and deployment visualization.
- Dataset Forge – upload assets, auto-balance modality ratios, and monitor quality cohesion.
- Auto Description Lab – generate narrative captions, prompt genomes, and Lumi Co-Pilot gap analysis.
- Model Playground – flip between Flux1.dev and Wan2.2 focus modes with adaptive hyperparameter hints.
- Training Conductor – visualize progress, losses, learning rates, and checkpoint cadence in real time.
- Pipeline Timeline – follow ingest → describe → train → deploy milestones with status highlights.
- Innovation Hub – signature extras like Persona Blend Engine, Audience Pulse, and Metric Hologram overlays.
npm install
npm run devOpen the provided local URL to explore the interface.
- Vite + React + TypeScript
- Tailwind CSS for the ultramodern neon aesthetic
- Framer Motion for smooth micro-interactions
- Zustand for lightweight state management
src/
components/ # UI modules for each training surface
data/ # Mock data used to simulate runtime telemetry
hooks/ # Global state store
styles/ # Tailwind entry point and global theming
This is a conceptual experience layer—hook up your own backend endpoints to connect real Flux & Wan training loops.