Write prompts that actually work

One simple framework for every model on PixenLabs — image, video, voice, music, and sound effects. Learn the formulas, the words that matter, and the difference between a weak prompt and a great one.

Start creating

Fundamentals

Six rules for every prompt

They apply whether you're generating an image, a film shot, a voiceover, a song, or a sound effect. Master these first.

Specific beats brief

Trade "a dog" for "a Shiba Inu with metallic plating leaping a rooftop gap at night." Detail is what separates generic from striking.

Put what matters first

Lead with your main subject, then layer the supporting detail behind it. What comes first sets the scene; trailing modifiers tend to get diluted.

Build it in layers

A reliable order works everywhere: subject → action → setting → composition → style. Decompose, don’t free-associate.

Describe in positive

Say "an empty street," not "no cars." Most modern models ignore negations unless they have a dedicated negative field.

Cut the filler

Drop "please," "amazing," "8k masterpiece." Empty adjectives dilute focus — concrete visual facts steer the model.

Iterate one change at a time

Start simple, then change a single variable per run. It’s how you learn what each word is actually doing.

Advanced

Techniques that level up any prompt

Lock the frame

State the aspect ratio as a number — 16:9, 2:3, 1:1 — and name the shot scale. It stops the composition from drifting and crops perfectly for wherever it lives.

Shoot it like a camera

Name a lens, aperture and light direction — “Woman in metallic trench coat inside a rainy phone booth, holding the receiver, eye contact. Rain-soaked Tokyo alley at night, red neon kanji reflections on wet asphalt, cyan fog. 35mm f/1.4, split red-cyan lighting, neo-noir cinematic.” Physical camera language is what gives real depth and believable light.