Writing the Subject: What's Actually In the Picture — Guides

One focal point

A picture is about one thing. Not one object; one point the eye lands on. A subject with two heroes splits the model's attention and you get two half-rendered ideas fighting over the center.

splita knight and a dragon and a burning village

focuseda knight walking away from a burning village, dragon shadow crossing the road

The second version still has all three ideas. But the knight owns the frame, the village becomes setting, and the dragon becomes one detail. That is a composition instead of an inventory.

Verbs make scenes

The single highest-leverage word in a subject is the verb. A noun gives the model a portrait; a verb gives it a moment. A fisherman stands and faces the camera. A fisherman mending a net has hands, posture, attention, and a reason for the light to fall where it does.

noun onlyan old botanist, greenhouse

with a verban old botanist repotting an orchid in a crowded greenhouse

Pick verbs a body can hold: reading, pouring, climbing, mending, waiting. Verbs of motion blur unless you want blur. Verbs of emotion (grieving, celebrating) work better as a pose or a mood word from the emotions wall.

Count what can be counted

Models are bad at implied numbers and good at stated ones. Some birds renders anywhere from two to forty. Three crows on a wire renders three crows, most of the time, and the miss is close. Numbers also anchor composition: three objects arrange themselves; several is a mob.

vagueships in a harbor at dawn

countedtwo fishing boats and one tall ship in a harbor at dawn

Anchor things in space

Prepositions are free composition control. Under, beside, in the doorway of, at the edge of: each one places the subject against the setting and tells the model where the empty space goes. Without an anchor, the subject floats in the center and the background wallpapers around it.

floatinga red bicycle, old town

anchoreda red bicycle leaning against a fountain at the edge of an empty square

Spend the words where they pay

A subject wants ten to twenty words: the focal point, its verb, one count, one anchor. Everything else belongs to other slots. Color goes to the colors wall, weather to the weather wall, texture to details. When the subject swells past twenty words, the model starts ignoring the middle of it.

everything crammed ina beautiful mysterious old lighthouse keeper with a weathered face and a yellow raincoat and a lantern standing dramatically on wet rocks during an epic storm at night with huge waves

budgeteda lighthouse keeper raising a lantern on wet rocks, storm waves behind. Then: yellow raincoat as a detail, night storm from the weather wall, one lighting word

Adjectives are the first thing to cut. Beautiful, mysterious, and epic carry no picture. Weathered face earns its place; it is something a camera can see.

Three rewrites, start to finish

beforecool cyberpunk street scene with lots of stuff happening

aftera noodle vendor serving a customer under dripping awnings, narrow alley, neon signs. Then: cyberpunk from the styles wall, sodium vapor from lighting

beforemagical forest with fairies and mushrooms and glowing lights everywhere

afterone fairy asleep on a mushroom cap, ring of smaller mushrooms around her. Then: bioluminescent as the detail, dappled light from the lighting wall

beforeepic dragon

aftera dragon coiled around a bell tower, townspeople watching from below. Then: one style word, golden hour, --ar 2:3 for the height

Where this fits

The subject is step one of the Start Here skeleton. The builder leaves the subject box empty on purpose; its dice deal nouns when the blank stares back, and this guide is how you turn a dealt noun into a scene. When the picture comes back wrong anyway, fix the subject before reaching for the negative prompt.

Published here first.

Writing the subject: what's in the picture