◈ THE CRIT
The Crit reads composition the way an artist does — mass, void, drift, radial gravity, tonal architecture — and tells you what your image is structurally doing, in plain language.
RUNS ENTIRELY IN YOUR BROWSER · YOUR IMAGE NEVER LEAVES YOUR MACHINE
The structural readout above is computed locally and is deterministic. Deep Crit sends the image plus the readout to Claude for a written one-on-one critique — the professor's voice on top of the instrument's numbers. Bring your own Anthropic API key; it's stored only in this browser and the image goes only to Anthropic.
Most image analysis asks what is pictured. The Crit asks how the picture is built. It comes from the Visual Thinking Lens (VTL) — a measurement framework developed by drawing from the masters and then testing generative models against the same geometry.
Edges and tonal weight become a mass field — the same thing your eye weighs before it ever names a subject. From it: placement drift, void ratio, cohesion, dispersion, peripheral pull, orientation.
Does the mass organize around its own center or the frame's? Generators settle into a known attractor: centered, radially even, frame-obedient. The Default Gravity Index says how deep in that basin your image sits.
The readout becomes paragraphs: what the image is structurally doing, what its loudest decision is, what it's giving up, and two or three concrete things to try. Honest, specific, teachable.
The same instrument serves two audiences: people who make images, and people who make image-makers.
Portfolio reviews cost $50–200. The Crit gives you the structural half of that conversation any time — where your mass sits, what your voids are doing, whether your composition is a choice or a default.
Semantic evals miss a whole defect class: compositional monoculture. Diffusion doesn't compose — it settles. The VTL kernel measures where your model settles: centered mass, dead voids, frame-locked radial structure, collapsed orientation.