v0.1.0 - Initial Release

Refocus v0.1.0 - Initial Release

ComfyUI custom nodes for the Genfocus generative refocusing pipeline.

Features

• Genfocus Nodes (diffusers-based)

  • GenfocusModelLoader: Load FLUX.1-dev with Genfocus LoRA adapters
  • GenfocusDeblur: Restore all-in-focus images from blurry/defocused inputs
  • GenfocusBokeh: Generate realistic depth-of-field effects with controllable focus
  • GenfocusGenerate: General-purpose generation with LoRA adapters

• DepthPro Nodes

  • DepthProModelLoader: Load Apple ml-depth-pro model
  • DepthProEstimate: High-quality metric depth estimation with visualization

• Depth Utilities

  • DepthMetricToRelative/Inverse: Convert between depth formats
  • FocalPXtoMM/FocalMMtoPX: Focal length conversions
  • ComputeDefocusMap: Generate defocus maps for bokeh effects
  • SelectFocusPoint/FocusPointFromMask: Interactive focus selection

Requirements

• Python 3.10+
• ComfyUI
• PyTorch with CUDA
• diffusers, transformers, accelerate, peft (for Genfocus nodes)
• ml-depth-pro (optional, for depth estimation)

Models Required

• FLUX.1-dev (diffusers format) - ~23GB
• Genfocus LoRAs (DeblurNet, BokehNet) - ~35MB
• DepthPro weights (optional) - ~500MB

See README for full installation instructions and example workflows.