Illustration Judgment
Illustration judgment is the accumulated discernment that separates a filter recipe that works from one that merely runs. Every filter string produces some output; the question is whether that output serves the illustration and the reader, and the answer cannot be derived from any specification. It is learned through looking — at many illustrations, across many themes, in many viewing contexts — and through the discipline of accepting that a recipe's apparent success may conceal a specific failure mode visible only to practitioners who have seen that failure before. The judgment is not transmissible by documentation; it is transmitted by mentorship, by working on real illustrations alongside someone who has cultivated it, by the accumulated experience of seeing what works and what doesn't.
In the AI Story
The judgment operates across several dimensions. A recipe can fail aesthetically (the illustration looks wrong against the background), functionally (the illustration is unreadable at a particular viewport size), or ethically (the illustration, through filter-induced distortion, misrepresents its subject in a way the original did not). The last is the most insidious: a hue rotation pushed too far can produce skin tones that carry implications the illustrator never chose, and the practitioner who cannot see that possibility will ship the mistake unaware.
The judgment also operates across time. An illustration that works in the current theme may fail when the site adds a new theme six months later. A recipe calibrated to today's source aesthetic may not suit the source aesthetic the practice evolves toward. The best practitioners think in terms of recipes that age well — that survive theme additions, source refinements, and changes in viewing context — rather than recipes that produce the best result for the current configuration alone.
AI assistance changes the economics but not the fundamentals. A language model can generate filter strings from natural-language descriptions, test them against source assets in rapid iteration, and surface variants the practitioner might not have considered. What the model cannot do is judge whether a variant is good. It can produce, very fast, many candidates; the selection among them remains the human's work. The practitioner who treats the model as an accelerant for candidate generation and retains the evaluative role produces better work than either human or model alone.
The accumulation of judgment is also the accumulation of geological understanding — the layered, embodied knowledge that comes from extended engagement with a material. Stipple has its properties; filters have theirs; the intersection is where the judgment lives. A practitioner who has rendered hundreds of illustrations through dozens of recipe variants develops intuitions that cannot be derived from specifications, only from the specific history of having looked, failed, and looked again.
Origin
Illustration judgment is a contemporary instance of a much older craft virtue. The Renaissance humanists called it iudicium and considered it the highest intellectual skill. The wiki's application of the term to filter recipes is frank about the analogy: the tools are different, the faculty is the same.
Key Ideas
Multiple dimensions. A filter recipe can succeed aesthetically while failing functionally or ethically; judgment must operate across all three.
Time-aware. Recipes that work today may fail tomorrow as themes, sources, or viewing contexts evolve.
Not transmissible by documentation. Judgment is learned through mentorship and repeated engagement, not from specifications.
AI accelerates generation, not selection. Models can produce many candidate recipes quickly; the evaluative role remains with the practitioner.
Debates & Critiques
Can illustration judgment be automated? Proponents argue that sufficiently large models trained on curated illustration corpora will eventually select as well as human practitioners. Skeptics — including the author — argue that selection requires stakes, context, and a biography the model does not possess. The debate will be settled, as most such debates are, by evidence that accumulates slowly over years rather than by argument.
Appears in the Orange Pill Cycle
Further reading
- Ann Blair, Too Much to Know: Managing Scholarly Information Before the Modern Age (Yale University Press, 2010) — on iudicium as a cultivated faculty.
- David Pye, The Nature and Art of Workmanship (Cambridge University Press, 1968).
- Edward Tufte, Beautiful Evidence (Graphics Press, 2006).
- Eugene Gendlin, Focusing (Everest House, 1978) — on the felt sense as an evaluative capacity.