← Back to blog
AI Tools & Tutorials

Riya Goel · 27 August 2026

50 Google Flow Camera Commands to Turn Any Photo Into a Cinematic Shot

50 Google Flow Camera Commands to Turn Any Photo Into a Cinematic Shot

The short answer. Google Flow camera commands are short motion instructions you type next to your image prompt. Upload your photo, then add one command like /dollyin, /crane or /hyperlapse. Flow reads it as the camera move for that shot and animates the still along that path. One command per shot, three to five seconds, then cut.

Most people generate their first Flow clip, look at it, and quietly decide the model is overrated. The frame drifts a little, nothing really happens, and it looks like a photo with a slow pan on top. That is not the model being weak. That is a shot with no camera move in it.

A director never says "make it cinematic". They say "we push in on her, four seconds, then cut". The camera command is that decision, written down. Leave it out and Flow picks an average drift for you, and average drift is exactly what makes a clip read as AI at a glance.

How to run a camera command in Google Flow

  1. Open Flow at labs.google/fx/tools/flow and sign in with your Google account.
  2. Start a project and use Frames to Video, then upload your photo as the opening frame.
  3. Write the scene in plain English: what is in the frame, and what the light is doing.
  4. Add one command from the list below on the same line as the description.
  5. Generate, watch it once, and if the move is wrong change only the command. Change one thing at a time or you learn nothing from the result.

The prompt shape that works is short and boring, which is the point:

[what is in the frame] + [what the light is doing] + [/command] + [how long]
A steel water bottle on a wet cafe table, morning light from the left /dollyin 4 seconds

Before you start testing. Every generation spends credits. The free tier gives you 50 a day, Google AI Plus is $4.99 a month for 200 monthly credits, Google AI Pro is $19.99 for 1,000, and Google AI Ultra runs $99.99 to $199.99 for 10,000 to 25,000. Trying five camera commands on one photo is five generations. Decide the move before you generate, not after.

Push and pull: the six Google Flow camera commands you will use most

These change the distance between the camera and the subject. If you only ever learn six commands, learn these.

1. /zoom

Changes how much of the scene fits in the frame by adjusting the lens, while the camera itself stays where it is. Use it as a first guess when you are not sure whether you want to go in or out, then swap it for /zoomin or /zoomout once you have seen a result.

A ceramic coffee cup on a linen tablecloth, steam rising /zoom 4 seconds

2. /zoomout

Widens the lens so more of the surroundings appear around your subject. It is the reveal: the product first, then the room it lives in.

A perfume bottle on a marble counter, warm window light /zoomout 5 seconds

3. /zoomin

Tightens the lens onto a smaller part of the frame. Good for landing on one detail at the end of a shot, a logo, a label, an eye.

A leather wallet on a wooden desk /zoomin on the stitching 4 seconds

4. /dolly

Moves the whole camera through the scene instead of changing the lens. The background shifts against the subject, and that shift is what makes a clip read as filmed rather than scaled.

A pair of running shoes on a garage floor, single overhead bulb /dolly 4 seconds

5. /dollyin

Walks the camera towards the subject at a steady pace. This is the most useful command in the list, because it adds weight to a still image while asking the model to invent almost nothing.

A skincare jar on a bathroom shelf, soft daylight /dollyin slow, 4 seconds

6. /dollyout

Walks the camera backwards, away from the subject. Use it to end a sequence, because pulling away reads as a full stop.

A founder at a desk in an empty office at night /dollyout 5 seconds

Zoom or dolly. A zoom changes the lens while the camera stands still, so the scene flattens as it tightens. A dolly moves the camera itself, so near objects slide past far ones. If your clip looks like a photograph being scaled up, you asked for a zoom when you wanted a dolly.

Around and across

These keep the camera at roughly the same distance and change the angle you see the subject from.

7. /orbit

Circles the camera part of the way around the subject while keeping it centred. Best on a single object with clear space around it.

A watch on a black stand, rim light from behind /orbit 5 seconds

8. /orbit360

Completes a full circle and returns to where it started. Ask for it only on simple objects, because the model has to invent the side of your subject that the photo never showed.

A sneaker on a plain grey podium /orbit360 6 seconds

9. /pan

Swivels the camera left or right from a fixed spot, like turning your head. Always add the direction, because Flow will otherwise pick one for you.

A long dining table set for dinner, candlelight /pan left to right 5 seconds

10. /tilt

Swivels the camera up or down from a fixed spot. The standard use is bottom to top on anything tall, so the height actually registers.

A glass office tower at dusk /tilt up 5 seconds

11. /truck

Slides the whole camera sideways while it keeps facing the same way. Not the same as a pan: here the viewpoint travels, so foreground objects sweep past the lens.

A row of paint cans on a workshop bench /truck right 4 seconds

12. /pedestal

Raises or lowers the whole camera vertically while it keeps facing the same way. Use it to move from a tabletop view up to eye level in one clean motion.

A cake on a kitchen island, window light /pedestal up 4 seconds

Up in the air

Height reads as production budget. These are the commands that make a small shoot look like a big one.

13. /crane

Lifts the camera up and forward on an arm, so the shot rises and travels at the same time. The classic way to open on a location.

A courtyard restaurant with string lights, evening /crane 6 seconds

14. /drone

Generic aerial motion above the scene. Use it when you want height but do not care about the exact flight path.

A winding coastal road, late afternoon /drone 5 seconds

15. /dronefly

Flies the camera forward over or through the scene at a steady altitude. The workhorse for landscape and property footage.

A tea estate on a hillside, morning mist /dronefly 6 seconds

16. /droneorbit

Circles the scene from above with the subject held in the middle. This is the standard establishing shot for real estate and hotels.

A white villa with a blue pool, midday sun /droneorbit 6 seconds

17. /aerial

Holds a high, wide view looking down, with only gentle drift. Use it for context, a city, a coastline, a site, before you cut to anything closer.

A dense city block at golden hour /aerial 5 seconds

18. /fpv

First person view flying: fast, low and close to obstacles, the style you know from racing drone videos. It brings energy and it distorts detail, so keep it away from product close-ups.

A skate park at sunset, empty bowls /fpv 4 seconds

19. /fpvdrone

The same style with the drone named explicitly, which tends to push it towards a longer, swooping flight path. Use it when /fpv comes back too tame.

A mountain road with switchbacks, clear sky /fpvdrone 5 seconds

Following a subject

These need something in the frame that can plausibly move. On a static product shot they do nothing.

20. /tracking

Keeps pace alongside a moving subject so it stays in the same part of the frame while the background streams past.

A woman walking through a market lane, morning /tracking 5 seconds

21. /follow

Stays behind the subject and travels with it. It reads as pursuit, which suits journeys and reveals more than it suits products.

A cyclist on a wet street, headlights behind /follow 5 seconds

22. /handheld

Adds the small, irregular motion of a camera held by a person. This is the fastest way to stop an AI clip looking synthetic.

A cluttered studio desk with a half-finished sketch /handheld 4 seconds

23. /steadicam

Smooth walking motion with the bumps taken out, the look of a stabilised rig. Use it where /handheld feels too rough for the brand.

A hotel corridor leading to an open door /steadicam 5 seconds

Time and speed

These change how fast the world moves rather than where the camera goes.

24. /slowmo

Slows the action so small movement becomes visible. It only works if something in the frame genuinely moves: fabric, liquid, hair, dust.

Coffee pouring into a glass, backlit /slowmo 4 seconds

25. /timelapse

Compresses a long stretch of time into a few seconds with the camera holding still. Shifting light and cloud carry the shot.

A city skyline from a rooftop, clouds passing /timelapse 5 seconds

26. /hyperlapse

A timelapse where the camera also travels, so the scene speeds past and moves at once. This is the command that turns a flat photo into something with momentum.

A tree-lined street in autumn /hyperlapse 5 seconds

27. /speedramp

Starts at one speed and changes to another inside the same shot, usually fast and then suddenly slow. Use it to land on a single moment.

A basketball leaving a player's hands /speedramp fast to slow 4 seconds

28. /motionblur

Keeps the speed but smears the movement, so anything fast streaks. It hides detail, which is useful when the model is inventing detail badly.

Traffic crossing an intersection at night /motionblur 4 seconds

29. /shakycam

Deliberate instability, considerably stronger than handheld. Reserve it for tension, news-style footage, or anything meant to feel unsafe.

A corridor with a flickering tube light /shakycam 4 seconds

Whose eyes the shot belongs to

Point of view decides whether the viewer is watching the scene or standing inside it.

30. /pov

Puts the camera where a person's eyes would be, so the viewer becomes that person. Say whose eyes, or the model will guess.

Hands opening a parcel on a kitchen counter /pov of the customer 4 seconds

31. /firstperson

The same idea stated more strictly, and it usually keeps hands and forearms in frame. This is the unboxing, cooking and how-to shot.

Hands cutting vegetables on a wooden board /firstperson 5 seconds

32. /thirdperson

Places the camera just behind and above a subject, watching them over the shoulder. It is the game-trailer look, and it works well on walking shots.

A man walking into a lit doorway at night /thirdperson 5 seconds

Framing and focus

These decide how much of the subject you see and which part of it is sharp.

33. /closeup

Fills the frame with a small part of the subject. Pair it with very little camera motion, because a tight frame plus a big move is where AI video falls apart.

A hand fastening a watch strap /closeup 3 seconds

34. /wide

Pulls the framing out so the subject sits inside its full setting. The standard first shot of almost any sequence.

A lone farmhouse in a wheat field, overcast /wide 5 seconds

35. /macro

Extreme close focus on texture, the level where you can see grain, fibre and droplets. Best on real uploaded photos, because there is less for the model to invent.

A water droplet on a leaf /macro 3 seconds

36. /rackfocus

Shifts sharpness from one thing in the frame to another while the camera holds still. It directs the eye without moving anything at all.

A phone in the foreground, a person behind it /rackfocus front to back 4 seconds

37. /focuspull

The same shift described as an action rather than a state, which tends to make it slower and more deliberate. Use it when /rackfocus snaps too quickly.

A bottle in front, a bartender behind /focuspull 4 seconds

38. /whip

A very fast swivel that blurs everything in between. On its own it is a jolt, so most people use it as a cut point rather than a shot.

A neon sign on a wet street /whip 2 seconds

39. /whip-pan

The same fast swivel named as a pan, so it stays horizontal. This is the one to use when you want to cut two clips together on the blur.

A crowded night market stall /whip-pan left 2 seconds

Spins, freezes and loops

The showy ones. They earn attention when they are rare and cost you credits when they are not.

40. /spin

Rotates the camera around its own axis so the horizon turns with it. Strong effect, so keep it under three seconds.

A vinyl record on a turntable /spin 3 seconds

41. /rotation

Turns the subject rather than the camera, the look of a product on a turntable. Use it for anything you would show on an e-commerce page.

A ceramic mug on a white cyclorama /rotation 5 seconds

42. /bullet-time

Freezes the action while the camera keeps travelling around it. Expensive-looking and unreliable, so test it before you promise it to a client.

A dancer mid-jump in a bare studio /bullet-time 4 seconds

43. /freeze

Holds the frame still while something small continues, usually light or particles. Good under a title card.

A boxer's fist at the point of contact /freeze 3 seconds

44. /loop

Ends the clip where it started so it can play continuously with no visible join. Use it for anything sitting on a website or a screen behind a stall.

Steam rising from a bowl of noodles /loop 4 seconds

45. /reverse

Plays the motion backwards. Most useful as a trick shot: a splash pulling itself back into the glass.

Petals falling onto a table /reverse 4 seconds

Transitions and finish

The last five are not really moves. They decide how a shot ends and what it looks like when it does.

46. /morph

Changes one thing in the frame into another during the shot. The least predictable command here, so budget several attempts for it.

A pencil sketch of a chair becoming the finished chair /morph 5 seconds

47. /transition

Asks for a deliberate exit from the shot rather than a hard stop, so it joins cleanly to the next clip. Say what it should pass through: light, a wall, a hand.

A door closing towards the lens /transition through the darkness 3 seconds

48. /glitch

Adds digital tearing, dropped frames and colour shifts. It suits tech and music content, and it destroys anything that needs to look premium.

A CRT monitor showing a login screen /glitch 3 seconds

49. /cinematic

A catch-all that pushes the shot towards film: shallower focus, a wider aspect, a graded look, a slower move. Use it on top of a real command, not instead of one.

A man reading by a single lamp /dollyin /cinematic 5 seconds

50. /epicshot

The biggest version of whatever the scene allows, usually a rising wide with dramatic light. It is the trailer ending, and it only works when the frame has some scale to give.

A fort on a ridge at sunrise /epicshot 6 seconds

Common mistakes when using camera commands in Google Flow in 2026

  • Asking for two moves in one shot. A dolly that becomes an orbit that becomes a tilt collapses into mush every time. One movement command per clip.
  • Leaving the direction off /pan, /tilt, /truck and /pedestal. Flow will choose one, and it is rarely the one you had in mind.
  • Running /orbit360 on a person or a detailed product. The model has to invent the side you never photographed, and it shows.
  • Putting a big travelling move on a close-up. Tight frame plus fast motion is where faces and hands break first.
  • Starting from a soft or low-resolution photo. A camera move magnifies whatever was already blurry.
  • Editing the scene description and the command in the same attempt, so you cannot tell which change caused the difference.
  • Treating /cinematic or /epicshot as the whole instruction. They style a shot, they do not choose a move for it.

FAQ

Do I need a paid plan to use camera commands in Google Flow?

No. The free tier gives you 50 credits a day and the commands work on it. Paid plans buy you more attempts, not more commands: Google AI Plus at $4.99 a month for 200 monthly credits, Google AI Pro at $19.99 for 1,000, and Google AI Ultra at $99.99 to $199.99 for 10,000 to 25,000.

Can I use more than one camera command in a single shot?

Stick to one movement command per clip. You can add a look command like /cinematic on top of it, but two movements in the same clip usually collapse into drift. If you need a rise and then a push, make two clips and cut them together.

What is the difference between /zoomin and /dollyin?

/zoomin changes the lens while the camera stays put, so the scene flattens as it tightens. /dollyin moves the camera through the space, so nearer objects pass the farther ones. One looks like a photograph being scaled up, the other looks like film.

Why does my clip seem to ignore the camera command?

Usually the prompt is fighting it. A long description with several actions in it leaves the model nothing left to spend on the move. Cut the description back to one subject and one lighting condition, keep the command on the same line, and keep the clip to three to five seconds.

Which command should I learn first?

/dollyin. It works on almost any uploaded photo, it asks the model to invent the least, and it is the single move that makes a still read as a shot instead of a slideshow.

In short

Flow will animate your photo either way. The camera command is what decides whether the result looks like a film clip or a screensaver, so pick the move before you spend the credit: /dollyin for weight, /crane for arrival, /hyperlapse for momentum, /rackfocus for attention. Give it a direction where the move has one, hold it to one movement per clip and three to five seconds, then cut. Fifty commands looks like a lot to memorise, but you will do most of your work with about eight of them.

TagsGoogle FlowVeo 3.1AI videocamera movementAI filmmakingprompt engineering

Need GTM creative shipped at AI speed?

One team for content, decks, landing pages, SEO/GEO, and video.

Get started