Featured

HomeStaging AI

AIContext: Kandy / Kandixi
Stable DiffusionControlNetPyTorchDockerGPUPython
HomeStaging

Distributed GPU worker for AI-powered image processing, orchestrating multiple models: Stable Diffusion 1.5 (RealisticVision), ControlNet (depth-guided), DepthPro (Apple), Segment Anything (Meta).


3 main features:

  • Generation:: interior scenes from a source image, depth map guidance, 10 design styles × 5 room types, catalog of 250+ prompts
  • Inpainting:: selective modification of zones (walls, floors, ceiling) with complete pipeline (segmentation → mask → color filtering → inpainting → blending)
  • Upscaling:: 2× to 4× enlargement

  • Architecture: Multi-worker stateless with PostgreSQL locking, resilience (retry/backoff), Docker GPU containerization (NVIDIA CUDA 12.8).

    Tech Stack

    PythonPyTorchCUDA 12.8Stable DiffusionControlNetDepthProSAMDockerPostgreSQL