Realistic Luma video comes from one clear structure written in plain English, not from stacking adjectives or director codes. Describe a single subject doing one action in a defined scene, name one camera movement, and set the style. Luma Ray3 is a reasoning-driven model in Dream Machine, so it reads every word literally — be specific, avoid contradictions, and keep to one main action. This guide breaks down the five-part formula, builds a prompt step by step, and gives 10 paste-ready examples.

Advertisement

The 5-part Luma formula

Realistic shots follow this order: Subject + Action + Scene + Camera movement + Style/Mood. Everything is written as natural language — no brackets, no director syntax. Ray3 parses full sentences, so write the way you would describe a shot to a cinematographer.

PartWhat it doesExample fragment
SubjectWho or what, with concrete visual detaila weathered fisherman in a yellow oilskin coat, grey stubble
ActionOne main action, with material and timingslowly hauling a wet rope hand over hand, fabric draping with weight
SceneThe place that grounds the subjecton the deck of a small trawler, open grey sea behind him
Camera movementOne plain-English move that complements the actionslow dolly in to a medium shot
Style / MoodGrade, tone, and reference in a few wordsdocumentary, muted color, overcast morning light

Because Ray3 reasons over the whole prompt, physics reads best when you spell out materials and timing: say "steam rising naturally" or "fabric draping with weight," and use pacing words like "slowly" or "rapidly." English is the most stable input language.

For faster iteration, keep the formula in a reusable structure. Our Luma prompt cheat sheet and Luma prompt templates lay out fill-in-the-blank versions of each part.

A worked example

Build the prompt one part at a time so each element earns its place.

1. Subject. Start with a specific person and a couple of visual anchors: "A weathered fisherman in a yellow oilskin coat with grey stubble."

2. Action. Add one main action, with material and timing so the physics resolves cleanly: "slowly hauls a wet rope hand over hand, the coat fabric draping with weight."

3. Scene. Ground it in a real place: "on the deck of a small trawler, open grey sea behind him."

4. Camera movement. Name one move in plain words that complements the action: "Slow dolly in to a medium shot."

5. Style / Mood. Close with grade and light: "Overcast morning light, soft and cool. Documentary, muted color."

Put together: "A weathered fisherman in a yellow oilskin coat with grey stubble slowly hauls a wet rope hand over hand on the deck of a small trawler, the coat fabric draping with weight, open grey sea behind him. Slow dolly in to a medium shot. Overcast morning light, soft and cool. Documentary, muted color." That is the whole prompt — one action, one move, one clear mood. Set duration, aspect ratio, and resolution in the panel, not in the text.

10 example prompts

Each prompt below is complete and paste-ready. Drop it into Dream Machine, keep the camera move in plain language, and set duration and aspect ratio in the panel.

1. Fisherman on deck (b-roll)

A weathered fisherman in a yellow oilskin coat with grey stubble slowly hauls a wet rope hand over hand on the deck of a small trawler, the coat fabric draping with weight, open grey sea behind him. Slow dolly in to a medium shot. Overcast morning light, soft and cool. Documentary, muted color.

Why it works: one action plus a single slow dolly keeps skin and wet fabric physics believable. (16:9, 1080p, 10s, text-to-video)

2. Barista pouring latte art (product)

A young barista in a black apron carefully pours steamed milk into a ceramic cup, forming a leaf pattern, at a marble cafe counter with a chrome espresso machine behind. Static locked close-up on the cup and hands, steam rising naturally. Warm window light from the left, soft shadows. Calm, artisanal, natural color.

Why it works: a locked frame plus one delicate action keeps liquid physics clean and avoids hand warping. (9:16, 1080p, 5s, text-to-video)

3. Runner at dawn (action)

A woman in grey running gear jogs steadily along an empty coastal boardwalk, ocean to her right. Tracking shot alongside her at a wide framing. Golden dawn light with long shadows and gentle lens flare. Fresh, energetic, cinematic.

Why it works: a tracking move matched to the subject's motion complements rather than fights it, so it reads smooth. (16:9, 1080p, 10s, text-to-video)

4. Chef flipping vegetables (action)

A chef in a white jacket rapidly tosses sliced vegetables in a steel wok over a gas flame in a busy restaurant kitchen, steam and a bright flame flare rising. Handheld push-in to a medium close-up. Warm tungsten light. Fast, appetizing, energetic.

Why it works: the handheld push-in adds life without a second move, and one flame source anchors the lighting. (16:9, 1080p, 5s, text-to-video)

5. Vintage car on a mountain road (nature)

A red 1960s convertible drives along a winding mountain road with pine forest on both sides. Crane up to reveal the valley below. Late-afternoon sun, warm and hazy with soft atmosphere. Nostalgic, cinematic.

Why it works: one revealing crane move builds scale while the car does the only action. (21:9, 1080p, 10s, text-to-video)

6. Rain on a city window (b-roll)

Rain streaks slide down a cafe window with a blurred neon street beyond, a coffee cup steaming naturally on the sill in the foreground. Static locked close-up, shallow depth of field. Cool blue night light with warm neon reflections. Moody, quiet, teal-and-orange grade.

Why it works: a locked frame and shallow focus let the rain and steam carry realistic micro-motion. (16:9, 1080p, 5s, text-to-video)

7. Product hero: sneaker turntable (product)

A white leather sneaker sits on a matte grey pedestal in a studio. Slow orbit 180 degrees clockwise around the shoe. Soft, even softbox lighting with a subtle gradient background. Clean, premium, minimal.

Why it works: a single slow orbit on a still subject is the safest realistic product move; start from a product photo in image-to-video for the cleanest result. (1:1, 1080p, 5s, image-to-video)

8. Child blowing bubbles (character)

A small child in a striped shirt gently blows a stream of soap bubbles in a sunlit backyard, green grass and a wooden fence behind, bubbles drifting slowly. Slow pull back to a medium shot. Bright afternoon sunlight, warm and soft with bokeh. Joyful, tender, nostalgic.

Why it works: one gentle action and a single slow pull back keep the face stable and expressive. (16:9, 1080p, 5s, text-to-video)

9. Blacksmith at the forge (character)

A muscular blacksmith hammers a glowing orange blade on an anvil in a dark stone workshop, sparks flying rapidly. Static locked medium close-up. Firelight from the forge, deep shadows with a strong orange glow on the face. Intense, gritty, dramatic.

Why it works: firelight as the only source plus one repeating action produces convincing sparks and skin. (16:9, 1080p, 10s, text-to-video)

10. Astronaut looking out a window (vertical)

An astronaut in a white suit slowly turns to look out a spacecraft window at Earth below. Slow dolly in toward the visor. Cool blue Earthlight on the helmet with soft interior fill. Awe, calm, cinematic.

Why it works: a slow dolly in and one subtle turn read as weightless and real; the 9:16 frame suits vertical feeds. (9:16, 1080p, 5s, text-to-video)

Advertisement

Ray3 features for realism

Ray3 (latest Ray3.14) ships several tools that raise realism and cut iteration cost. Learn them, then browse the best Luma prompts roundup for ready-made shots.

Draft Mode → Hi-Fi. Draft renders roughly five times faster and for about five times fewer credits. Iterate variations in Draft, pick the take you want, then press Upgrade to Hi-Fi for the final quality render.

Keyframes. Set a start frame and an end frame and Ray3 interpolates between them, giving you controlled, predictable motion instead of a guess.

Native HDR. Ray3 renders in 16-bit HDR (ACES EXR), so highlights and shadows hold detail for a graded, filmic look straight out of the model.

Image-to-video. Upload a still and add a short motion prompt. The composition and subject look are preserved from your frame, which is the most reliable path to realism.

Extend and Loop. Push a shot to about 30 seconds with Extend, or use Loop to build a seamless repeating clip for backgrounds and b-roll.

Modify. Restyle existing footage while preserving its motion, in about 18 seconds — useful for changing grade or look without regenerating the shot. For consistent characters across shots, Character Reference locks a subject's appearance.

Common mistakes to avoid

Most unrealistic results trace back to a handful of habits. Fix these first, then compare cinematic setups in our Luma cinematic b-roll prompts.

Contradictory instructions. Ray3 reads every word literally, so "static locked shot" plus "fast orbit" fights itself and produces jitter. Keep the camera direction internally consistent.

Stacking multiple actions. "Walks, waves, and sits down" forces the model to guess timing and it warps. Keep one main action per shot.

Vague verbs. "Moves" or "does something" gives the model nothing to reason about. Use precise verbs with timing, such as "slowly hauls" or "rapidly tosses."

Fighting the composition in image-to-video. The uploaded still already sets the look. Prompt only the motion and camera; re-describing the scene fights the reference and introduces artifacts.

Over-long prompts. Because Ray3 weighs every word, an over-stuffed prompt buries the main action. Trim to two or three tight sentences that cover the five parts.

Expecting audio. Ray3 outputs silent video. Describing sounds wastes words; add music and effects in an editor afterward.

Frequently Asked Questions

How long should a Luma prompt be?

Long enough to cover the five parts and no longer. Two or three natural-language sentences with a single main action, a defined scene, and one camera move is the sweet spot. Ray3 reads every word literally, so extra clauses dilute the main action and invite drift.

How do I control the camera in Luma?

Write the camera move in plain English, not brackets or director codes. Phrases like slow dolly in, orbit 180 degrees clockwise, or tracking shot alongside work well. Make the camera motion complement the subject's motion rather than fight it, and keep it to one move per shot for realism.

Does Luma generate audio?

No. Ray3 outputs silent video. Add music, dialogue, or sound effects in a separate editor afterward. Do not spend prompt words describing sounds, since the model ignores them.

What is Draft Mode and when should I use it?

Draft Mode renders roughly five times faster and for about five times fewer credits, so it is built for iteration. Explore variations in Draft, then choose the best take and press Upgrade to Hi-Fi to render the final version at full quality.

Text-to-video or image-to-video for realism?

Image-to-video gives the most reliable realism because your uploaded still locks the composition and subject look. In that mode, upload the frame and write only a short motion prompt. Use text-to-video when you have no usable reference image.

How long can a Luma clip be?

Text-to-video renders 5 or 10 seconds, and image-to-video renders 5 seconds. You can push a shot to about 30 seconds with Extend, or use Loop for a seamless repeating clip.

How do I keep a character consistent across shots?

Use Character Reference, which locks a subject's look across generations. Pair it with Keyframes for controlled start-to-end motion, and keep each shot to one main action so the face and body stay stable.

Advertisement