Graded valley landscape

RAW in.
Masterpiece out.

A local agent reads your photo and forges it — HDR, editorial grade, face restoration, retouch, depth, deblur — fully offline. Nothing is ever uploaded.

Get the course See it work A whole toolbox · 3 ways to run · Mac / Win / Linux
Scroll
RAW · untouchedForged
CoastlineDrama look · wide dynamic range
RAW · untouchedForged
AuroraEpic look · low-light lift
RAW · untouchedForged
ReflectionEpic look · wide dynamic range
RAW · untouchedForged
DunesDrama look · deep shadow recovery
02 — One photo, ten moods

Ten looks.
One word.

Name the mood — museum, vivid, noir, gritty — and the agent renders it. The same coast, ten ways. Drag to scroll →

Natural
Natural01
Vivid
Vivid02
Museum
Museum03
Soft
Soft04
Hyperreal
Hyperreal05
Epic
Epic06
Drama
Drama07
Vintage
Vintage08
Noir
Noir09
Gritty
Gritty10
03 — Far beyond grading

It's a whole
darkroom.

HDR is just the start. The same agent restores faces, retouches skin, deblurs, adds motion and upscales — drag to compare.

DegradedRestored
Magazine restoreFace restoration + editorial grade
RAWRetouched
Skin retouchSmooth · de-shine · blemish removal
SoftSharpened
DeblurRecover edges & detail
StillIn motion
Motion blurAdd cinematic speed
Upscaled detail 4× upscale
low-res input300 px in
AI upscaleReal-ESRGAN · 300 px → 1200 px
Portrait bokeh Bokeh
Portrait bokehSubject sharp · background melted
04 — It reads the scene

Depth, light
& weather.

On-device depth estimation, atmospheric synthesis and cinematic palette transfer — no cloud, no GPU required.

Depth map of a city skyline
Depth, understoodMonocular depth estimation · runs on-device
Weather
AWeather

Synthetic fog · rain · sun

Palette
BPalette

Cinematic color transfer

Tilt-shift
CTilt-shift

Selective focus planes

06 — How it runs

Three ways
to run it.

One catalog, three postures. Your privacy, GPU and workflow decide which.

Sovereign

Fully offline

A local model is the brain; your hardware does the compute. Images never touch a network — built for privacy, defense and air-gapped work.

Hybrid

Agent + local tools

Drive the catalog from an agent host (Claude Code, Cursor, …) over MCP. A frontier brain reasons; the pixels stay on your machine.

Cloud

Burst to scale

No local GPU? Run the exact same filters on hosted compute for the heaviest models — same results, no hardware.

07 — The build

This page is the teaser.
The build is the course.

You've seen the results. What you don't get here is the how — the registry, the model wiring, the prompts that let one small local model orchestrate this whole toolbox without ever guessing at pixels.

That's the course: build a sovereign photo agent from zero — local model, permissively-licensed AI models, the full catalog — running sovereign, hybrid and cloud, on Mac, Windows and Linux, with or without a GPU.

One-time payment · instant repository download + lifetime course access.

Shown hereThe results
In the courseThe architecture
In the courseThe model wiring
In the courseThe prompts
In the courseThe full catalog
In the courseCross-platform setup
What you get
  • The full video masterclass — every lesson, from first principles to shipping
  • The complete reference repository — filter registry, HDR engine, agent, providers and MCP server — yours to keep, study and build on
  • Every look and effect shown on this page, with the recipes behind them
  • All three deployment paths — sovereign, hybrid and cloud
What you need
  • Mac, Windows or Linux — no GPU required for the classical toolbox
  • Python, and a local LLM (Ollama or any OpenAI-compatible server)
  • A GPU (CUDA / ROCm / Apple Silicon) only for the heavier AI models
  • Comfort with the command line — this is a builder's course, not point-and-click