Starting a website?

Purpose-Built Templates

ZU2B Blueprints

Business in a Box

Reference

ZU2B Pulse™

See what visitors experience before they leave — paste any public URL and get a plain-English performance report in seconds.

Scan Your URL Free
FAQ • Support

Why an AI Image Tool Ignores Your Lighting Instructions

Field Note #132
🔬 Generalized 🟡 Intermediate ⏱ 30 Minutes ✓ Observation from Experience
You ask an AI image generator for “warm lighting” on a character, or “the subject slightly out of focus.” You get back a result that technically matches the words, but not the effect you actually wanted — the lighting barely changed, or nothing is actually blurred at all. You try rewording it, emphasizing the same idea more strongly, and get a similarly underwhelming result.

What’s actually happening
Vague, descriptive, mood-based language (“warm,” “soft,” “slightly”) leaves a lot of room for interpretation, and image models tend to interpret it conservatively or inconsistently — sometimes folding the instruction into an unrelated part of the image (like shifting eye color instead of adding rim lighting), sometimes barely acting on it at all. Specific, technical, or numeric language gives the model far less room to interpret loosely, because it maps more directly onto a concrete visual effect.

Two concrete examples from real attempts:
“Warm rim light on the character” produced no visible effect across two attempts. “Warm rim light, approximately #FFB366, along the edges of the chassis” — naming an exact hex color — worked on the very next attempt.
“Slightly out of focus” produced no actual lens blur at all; the model interpreted it as a mood/lighting note rather than a literal depth-of-field instruction. The more precise photographic term, “shallow depth of field, [subject] soft/blurred in the foreground, [background] in sharp focus,” is far more likely to trigger genuine blur rendering, since it names an actual camera/rendering concept instead of a vague impression.

How to tell the difference
If a prompt uses subjective, mood-based adjectives (“warm,” “soft,” “moody,” “dramatic”) for an effect that matters to you specifically, that’s a signal it’s likely to be under-realized or applied somewhere unintended. If the same idea can instead be expressed as a specific value (a hex code, a numeric angle, a named photographic/rendering technique), that’s usually the more reliable path.

The fix
Translate the vague creative direction into the most concrete, specific version of itself before submitting the prompt:
Colors → exact hex values, not “warm” or “cool”
Blur/focus effects → the literal term “depth of field,” not “slightly out of focus” or “blurry”
Expressions/poses → precise physical descriptions (“fingers loosely spread, palm up”) rather than emotional labels (“a friendly gesture”)

How to avoid it next time
Before submitting a prompt, scan it for any word that’s describing a feeling rather than a visual mechanism. Each one is a candidate for translation into something concrete and specific — and it’s usually worth doing that translation up front, rather than waiting to see if the vaguer version happens to land.

Tagged in :

More from the field