Experience professional, non-destructive 32-bit floating-point RAW editing with real-time progressive 3D Bokeh raymarching, neural surface normals relighting, and modular GPU shader plugins.
Drag the interactive divider below to see how progressive 32-bit raymarching transforms flat sensor captures into rich cinematic portraits.
Designed from the ground up for high-dynamic-range camera sensors, computational photography researchers, and discerning portrait photographers.
Interactive multimodal co-pilot with deep Thinking reasoning. Features separated Studio Retouching Rules, instant One-Time Prompts (with auto-clear & quick chips), 50% dampened convergence, and AI Bokeh A/B verification.
Phase 1 rapid triage and Phase 2 Center 50% Crop optical blur inspection. Includes 1-Click Full Auto Batch grading with resilient 10s persistent retry and smart skip resume.
All calculations run on 32-bit floating-point buffers in ModernGL. Hardware trilinear mipmap pyramids eliminate sensel aliasing, moirΓ©, and downsampling noise on 45MP Canon CR3 / Sony ARW files.
True optical depth-of-field simulation using Monte Carlo Vogel spiral raymarching. Includes one-click 3D Canvas Focus Eyedropper, cat-eye vignetting, chromatic aberration, and polygonal iris blades.
Auto-slices exports into dedicated aspect subfolders (Original, Photo_iPad_4x5, Portrait_2x3, Square_1x1, Story_3x4, Cinematic_16x9) with 5β master curation, plus sub-50ms instant Fast Viewer browsing.
Targeted 8-band luminance and hue adjustments in perceptual OKLCH color space. Brighten deep shadows or tone skies without unwanted saturation shifts.
Hand-paint masks with hardware-accelerated stroke interpolation. Rapid mouse sweeps produce smooth, continuous lines with zero gaps or stepping.
Heterogeneous GPU acceleration for DepthAnythingV2. Batched FP16 TensorCore inference delivers ~12x faster depth maps (0.024s/tile) with DirectML and Safe Mode CPU fallback.
Built-in Model Context Protocol server (SSE & stdio on port 8000) allowing external AI coding assistants to inspect histograms, extract crops, and apply live adjustments.
Lumens Deep automatically extracts spatial depth and surface geometry, turning flat 2D sensor data into interactive 3D volumetric scenes.
The complete non-destructive render with custom OKLCH tone curves, color balance, and hardware-accelerated ModernGL layer accumulation.
Choose between vintage polygonal iris blades or ultra-smooth modern circular lenses.
5 Blades β Pentagon (Classic Hollywood Bokeh)
In Lumens Deep, slider position 0 creates a triangle, position 1 creates a square, ..., and position 10 delivers creamy circular bokeh.
Drag the control points on the curve to see live 32-bit linear GPU tone corrections applied to the RAW photo in real time!
π‘ Drag any of the 8 points up/down to shape the tone curve. Double-click to reset to 0 EV.
Sculpt colors in perceptual OKLCH color space. Selectively boost or mute saturation, adjust hue brightness, or shift color spectrums with 0% skin-tone artifacting.
π‘ Saturation Mode: Drag points to selectively boost or desaturate specific hues across the spectrum.
A modern, open-source computational photography workstation built for the next generation of visual artists.
| Feature / Capability | Lumens Deep | Adobe Lightroom | Darktable | Capture One |
|---|---|---|---|---|
| Non-Destructive 32-Bit Float Engine | Native ModernGL | ProPhoto | 32-bit float | Native |
| Progressive 3D Bokeh Raymarching | Full Optical GPU | 2D Blur Filter | None | None |
| Offline Neural 3D Normals & Depth | Built-in ONNX | Cloud/Limited | None | None |
| Aperture Blade Shape Geometry (3-12 Blades) | Configurable | Fixed | None | None |
| Perceptual OKLCH Tone & Color Equalizers | Native OKLCH | Standard HSL | Color Calibration | Standard HSL |
| Sub-Pixel GPU Brush Stroke Interpolation | Zero Gaps | Available | Stepped | Available |
| Extensible Python / GLSL Plugin System | 1-File Plugins | Closed | Lua Scripts | AppleScript only |
| Open Source & Free Forever | MIT License | Subscription | GPLv3 | $299+ |
Hi, Iβm Volodymyr Makovetskyi. As the Lead Developer of 3DCoat and creator of Scanimat technology (scanimat.com, scanimate.io), my journey in 3D computer graphics began back in 1995, with over 20 years of commercial engineering experience spanning from early 3D model architectures to high-end real-time GPU rendering engines.
I originally began building Lumens Deep as a personal hobby project to edit family photos with the quality, optical realism, and depth fidelity I always dreamed of. Traditional industry editors like Adobe Lightroom, Capture One, and Darktable are fantastic for flat 2D color grading, but none of them generate ultra-precise 40MP+ metric depth maps where every single pixel aligns perfectly with RAW sensor resolution. None of them allow you to treat a capture as a living 3D environment β isolating depth planes, relighting surfaces via high-frequency surface normals, or computing genuine optical bokeh raymarching with customizable aperture blade geometry.
Lumens Deep was purposefully engineered for the next era of AI Agent collaboration via the Model Context Protocol (MCP). Unlike generative diffusion tools that hallucinate and redraw over your photos, AI agents in Lumens Deep act as precision studio assistants β they analyze and adjust the native GPU sliders, OKLCH curves, and 3D geometric masks without ever painting over or compromising your photographβs genuine optical authenticity.
I am sharing Lumens Deep with the global photography and developer community as a 100% free and open-source project (MIT License). It is built to give visual artists total creative freedom, without subscriptions, walled gardens, or cloud dependencies.
Get started with Lumens Deep in less than 60 seconds on Windows, macOS, or Linux.
β‘ NVIDIA GPU Acceleration (~12x faster AI depth & normals):