Every prompt here is written to be pasted straight into Muse Video, in the five-clause shape that lets you revise one thing at a time. Each comes with a diagram of the shot it requests and a rating for how much coherence it demands because the prompts that ask less of the model are the ones that come back usable.
The thumbnails are diagrams we drew of the intended shot not frames of Meta’s output. We publish no invented results.
Each prompt is one sentence in five parts subject, style, motion, duration, audio colour-coded so you can see the seams. Keep them as separate clauses when you edit: that is what lets you fix the motion without disturbing the light, and it is the difference between learning the tool and rewriting prompts in circles.
Paste one in as written, swap the subject for yours, then change exactly one clause based on what comes back.
The difficulty rating is not about how good the clip will look. It is how much has to stay consistent from frame to frame: a held camera on a quiet scene asks almost nothing, while hands, faces, liquids, animals and anything crossing the frame ask the model to remember things. Hard prompts are worth attempting they just cost more attempts, which is the real currency.
“Change first” names the clause most likely to need adjusting on the second attempt, based on which dimension that kind of shot usually gets wrong.
Every thumbnail on this page is a diagram we drew: composition, subject placement, and an arrow for the direction of movement. None of them is a frame from Meta’s output, and none is a claim about what your clip will look like.
That is deliberate. A page that invents frames and presents them as results is doing the same thing as a page that invents clip lengths and resolutions and we already refuse to do that. The diagram tells you what you are asking for; only your own generation tells you what arrives.
The easiest thing generative video does well, and the right place to start on a new account. Light, air and steam move; nothing has to be remembered.
01
A shaft of morning sun falling across a wooden kitchen worktop with a ceramic jug, warm natural light, shallow depth of field, shot on 35mm film, camera completely still while dust motes drift upward through the beam, a short clip of a few seconds, quiet room tone, no music.
02
Raindrops running down a dark window pane with out-of-focus street lights behind, moody low-key lighting, deep blues and amber bokeh, camera locked off, only the water moving downward, a short clip, just a few seconds, soft rain and distant traffic, no music.
03
A shelf of worn hardback books with faded cloth spines, soft diffused window light, muted colour, documentary realism, camera pushing in very slowly and steadily, nothing else moving, a short clip of two or three seconds, silence with faint room ambience.
04
An empty café interior, chairs tucked under marble tables, blinds half open, cool morning light, desaturated palette, wide lens, camera drifting slowly to the left, no people and nothing else in motion, a short clip of a few seconds, quiet ambience, a fridge hum, no music.
05
A brass desk lamp glowing over an open notebook and a closed laptop, warm tungsten pool of light against a dim blue room, cinematic contrast, camera almost still, a barely perceptible drift right, light steady, a short clip of two or three seconds, faint evening traffic outside, no music.
Product shots are forgiving as long as the object stays put and the camera does the moving. Anything that turns, pours or splashes moves straight to hard.
06
A steel wristwatch resting on rough grey stone, hard directional light with crisp shadow, high-end product photography, camera tracking slowly sideways past it, watch perfectly still, a short clip of a few seconds, no music, a faint ticking.
07
A faceted glass perfume bottle on a dark reflective surface, studio lighting, deep shadows, jewellery-advertisement polish, bottle stationary while a single highlight sweeps slowly across the glass, a short clip, two or three seconds, no audio at all.
08
A single white trainer centred on a seamless pale grey backdrop, even soft studio light, no harsh shadow, catalogue look, shoe rotating slowly on a turntable, camera fixed, a short clip of a few seconds, no audio.
09
Dark coffee pouring into a clear glass on a pale counter, backlit so the liquid glows amber, macro food-advertisement style, camera locked off, only the stream and the rising level moving, a very short clip, two seconds, the sound of pouring liquid, no music.
10
An unbranded frosted glass cosmetic jar with condensation on its side, cool clinical light, extreme macro, shallow focus, jar still, one droplet sliding slowly down the glass, a very short clip of about two seconds, no audio.
Fog, wind, water and cloud are the medium’s natural home: continuous, formless movement where nothing needs to keep its identity from one second to the next.
11
Tall dark pines on a hillside with low fog between the trunks, flat overcast light, near-monochrome, cool and still, camera fixed while the fog drifts slowly left to right, a short clip of a few seconds, wind in branches, no music.
12
A thin wave washing up over dark wet pebbles on a shoreline, soft grey afternoon light, muted and natural, handheld documentary feel, camera looking down and held steady, only the water advancing and retreating, a short clip of a few seconds, water over stones, gulls far off, no music.
13
A field of ripe wheat under a pale sky, golden late-afternoon light, warm and slightly hazy, anamorphic look, camera still at crop height while a gust moves through the heads in waves, a short clip of a few seconds, wind and rustling stalks, no music.
14
Snow falling through the cone of a single amber streetlight on an empty road, high contrast night, warm light against cold blue dark, grainy, camera fixed, snow drifting down and slightly sideways, a short clip of a few seconds, muffled silence, faint wind, no music.
15
A bare mountain ridge with cloud passing behind it, cool blue hour palette, wide vista, crisp and clean, camera locked off, cloud moving steadily across at slightly accelerated speed, a short clip of a few seconds, high wind, nothing else, no music.
Hands, faces and animals are where generative video earns its reputation for uncanny results. These five are written to reduce the burden: partial framing, one simple action, no dialogue.
16
Two hands folding brown paper around a small box on a wooden table, seen from directly above, soft daylight, warm neutral tones, artisan brand film, camera fixed overhead, hands making one single folding motion, no faces in frame, a very short clip of about two seconds, paper crinkling, no music.
17
A cyclist crossing the frame in front of a plain painted wall, flat overcast light, bold flat colour, editorial stills feel, camera fixed and side-on, the rider entering and leaving frame once, distant and small, a very short clip, about two seconds, tyre hum on tarmac, no music.
18
A pair of hands cradling a chipped stoneware mug of tea, no face in frame, warm window light, cosy muted palette, shallow focus, hands and camera both still, only steam rising and curling, a short clip of two or three seconds, quiet room tone, no music.
19
A hand turning one page of a large open book on a desk, seen from above, soft even light, aged paper tones, quiet library mood, camera fixed overhead, one page lifting and settling, nothing else moving, a very short clip, about two seconds, paper turning, no music.
20
A tan short-haired dog lying on a wooden porch in shade, bright summer daylight just outside the shade, natural colour, home-video warmth, camera still and low, the dog resting with only its tail moving slowly, a short clip of two or three seconds, cicadas and a distant screen door, no music.
The most reliably usable category, and the most overlooked. Nothing in these has an identity to maintain, which is exactly why they come back clean and they solve real problems in an edit.
21
A drop of deep blue ink blooming outward through clear water, black background, high contrast, macro scientific look, camera fixed, the ink expanding slowly and evenly outward, a short clip of a few seconds, no audio.
22
An unbranded pink neon sign glowing above wet black asphalt with its reflection below, night, saturated magenta and cyan, cinematic anamorphic flare, camera still, the neon pulsing gently and the reflection rippling, a short clip of a few seconds, a faint electrical hum and light rain, no music.
23
A sheet of heavy cream watercolour paper filling the frame, no text on it, raking side light picking up the tooth of the paper, warm neutral, camera almost static with a very slow downward drift, the light shifting subtly, a short clip of a few seconds, no audio.
24
Thin coloured smoke rising and curling against a pure black background, magenta and teal lit from the side, high contrast, studio smoke test, camera fixed, smoke drifting upward and folding into itself, a short clip of a few seconds, no audio.
25
Soft overlapping blobs of violet, teal and amber filling the frame like a blurred gradient, heavily out of focus, no hard edges, calm and modern, camera static while the colours drift and blend continuously, ending close to how it began, a short clip of a few seconds, no audio.
These last five are the ones worth generating before you need them. A folder of clean abstract backgrounds and title plates solves problems in an edit at no notice, and they are the cheapest clips to get right see social videos for where they fit.
The structure is the reusable part. Swap the content and keep the shape.
The style, motion, duration and audio clauses are doing the work. Replace the mug with your product and leave the rest alone for the first attempt then adjust one clause. The revision loop.
Any hard prompt gets easier if the camera holds still and the subject does less. Turn a rotation into a drift, or an action into the moment just after it. That single edit converts most failures.
Generate the look as an image where iteration is cheap, then describe what you got in the style clause. Far more reliable than adjectives invented on the spot. Image prompts.
Sound is generated with the picture, so an unstated audio clause means inheriting whatever the model picks. “No audio” and “quiet room tone” are both real instructions. Audio in Muse Video.
No, and we say so plainly. They are diagrams we drew of the shot each prompt asks for composition and direction of movement. Publishing invented frames and calling them output would be the same sin as publishing invented specifications.
Because a video prompt carries subject, style, motion, duration and audio, and keeping them separate is what lets you fix one without disturbing the other four. It reads a little mechanically on purpose the reasoning.
How much has to stay consistent from frame to frame. Easy means a near-static scene with gentle camera movement; hard means hands, faces, liquids, animals or a subject crossing the frame the things that ask the model to remember something.
Yes, that is what they are for. Swap the subject for yours and leave the structure alone, then change one clause at a time based on what comes back.
Because a held or slowly drifting camera over a quiet scene asks the model to keep very little consistent, which is where generative video is strongest. Gentle movement is not a limitation to work around; it is the material why that is.
Say you want something short, in words rather than as a claimed specification. Meta has not published clip lengths, so the phrasing is a request rather than a setting you are dialling in.
Because sound is generated with the picture rather than added afterwards, so leaving it unsaid means inheriting whatever the model chooses. Asking for quiet room tone, or for no music, is a real instruction.
Name the clause it ignored and rewrite only that one. A full rewrite changes all five dimensions at once, so a fix and a regression arrive together and you learn nothing.
Any of the ones marked easy. They establish what your account actually produces length, dimensions, sound at the lowest cost in attempts, which is the unit you are spending.
Mostly, yes. The five-clause structure and the coherence-first thinking are properties of generative video rather than of one product, so they transfer with minor wording changes Muse Video vs Veo.
The prompts marked easy will tell you more about your account in fifteen minutes than any specification page can including every page on this site.
Independent guide not affiliated with Meta. The thumbnails are our own diagrams of the requested shot, not Meta output. Clip lengths, formats and limits are Meta’s to state; confirm them in Meta’s own documentation.