Lumiera

Lumiera — Next-Gen Local AI Suite for Video Generation, Character Swap & Image Synthesis

Welcome to the official itch.io page for Lumiera!

Lumiera is an all-in-one, portable workstation designed to run the most advanced generative AI models locally on your GPU — with complete data privacy, zero cloud subscription fees, and deep VRAM optimizations.

💻 System Requirements

⚠️ Important Note on Performance: The more VRAM and RAM you have, the significantly faster your generations will be.

• Minimum Specifications (Bare Minimum):
  • GPU: NVIDIA GPU with at least 8 GB VRAM
  • RAM: 16 GB System RAM
  • OS: Windows 10 / 11 (64-bit)
  • Generation Speed Notice: With 8 GB VRAM and 16 GB RAM, heavy model generation (video synthesis) is possible due to aggressive CPU offloading and quantization, but generation will take hours.
  • Disk Space: 100 GB+ free space on SSD / NVMe
  • Windows Pagefile (Swap): 32 GB – 48 GB (crucial to prevent crashes when loading heavy model weights)
• Recommended Specifications (Fast Generation):
  • GPU: NVIDIA RTX 3090 / 4080 / 4090 (16 GB – 24 GB VRAM)
  • RAM: 32 GB – 64 GB System RAM
  • Disk Space: 250+ GB free space on NVMe SSD (for storing models, checkpoints, and LoRA libraries)
  • Windows Pagefile (Swap): 48 GB – 64 GB+ on NVMe SSD

🧠 Supported Models & AI Architectures

🎬 AI Video Generation:

• Wan 2.1 & Wan 2.2: Wan2.1 T2V (1.3B & 14B), Wan2.1 I2V (14B), and WAMU Lightning architectures.
• LTX-Video (Lightricks): LTX-Video 0.9.7 & 2.3 (Safetensors and quantized GGUF models: Q4_K, Q6_K, Q2_K with integrated Audio VAE & Vocoder).

🎨 Image Generation & Turbo Models:

• SDXL & Stable Diffusion 1.5: Full support for custom checkpoints, LoRA adapters, and Civitai models.
• Krea-2 Turbo: Real-time ultra-fast diffusion generation.
• Z-Image Turbo & Juggernaut-Z: Next-gen high-speed image synthesis.

👤 Character Replacement & Face Tools:

• OmniGen 2 (BAAI): Multimodal character replacement and image conditioning.
• InsightFace & InSwapper: High-precision video & image face swap (buffalo_l, inswapper_128).
• IP-Adapter & FaceID SDXL: Identity-preserving image generation and character consistency.

✨ Upscaling & Restoration:

• Real-ESRGAN (x4plus): High-detail AI resolution upscaling.
• GFPGAN (v1.3 & v1.4): AI facial enhancement and detail restoration.
• U2Net: AI background removal and object segmentation.