The most useful thing to understand about image prompts is that attention is a fixed budget. The model divides its attention across everything you wrote. Whatever you spend the most words on becomes the most of the picture.

This surprises people, because it's the opposite of how writing usually works. In prose, adding detail about a necklace adds a necklace to the scene. In an image prompt, adding three sentences about a necklace tends to give you a photograph of a necklace.

The rule

The share of the frame a thing occupies roughly tracks the share of the prompt it occupies.

Describe eyes across four clauses and you'll get a close-up crop of a face. Describe a room in one clause and a character in six, and the room becomes a blurred backdrop.

What this means in practice

Describe the frame you want, not everything you can think of. A common instinct is to write everything known about a character into every prompt. That fights you: the more exhaustive the character description, the tighter the model crops to fit it all in.

Balance deliberately. If you want a character standing in a marketplace and you want to see the marketplace, the marketplace needs comparable prompt real estate. Six words of setting against sixty words of character gets you a character with a smudge behind them.

Cut before you add. If an image is too tight, the fix is usually deleting facial detail rather than adding more words. Framing itself is a separate prompt — see Framing & Shot Types.

Position matters too — but you don't control it

Weight isn't only about word count. Earlier words carry more influence than later ones in most architectures, and a detail at the end of a long prompt may barely register.

In DreamRunner the overall order is fixed — physical description comes before outfit, and the framing and background sit after both. You can't move a field, so the lever you have is order within a field: put what matters most at the start of the description, not buried three clauses in.

Note: this is also why the physical description quietly outranks the outfit. If a character's clothing keeps losing to their appearance, shortening the physical description works better than lengthening the outfit.

What about weighting syntax?

Some image tools accept numeric emphasis like (silver necklace:1.2). It is worth knowing that this is blunt, that prose-based styles largely ignore it, and that DreamRunner's own descriptions never use it.

Rebalancing the description is more reliable than tuning weights on top of an unbalanced one.

Note: if you find yourself stacking weights to force a result, the description is usually fighting itself. Rewrite it shorter before you tune it harder.

A worked example

Too tight — 80% of the words are face:

A woman with almond-shaped emerald eyes flecked with gold, long dark lashes, high cheekbones, a small scar through one eyebrow, full lips, and freckles across her nose, in a marketplace.

Balanced:

A woman with dark hair and green eyes walks through a crowded spice marketplace at midday. Stalls of red and ochre powder line both sides, awnings overhead, dust in the sunlight. Full-body shot from a few paces back.

Same character. The second one is a scene; the first is a portrait wearing a marketplace as a hat.