Flow AI shot prompts you can copy
Six shot setups written in the vocabulary Veo 3.1 responds to, with the reasoning behind each choice and the failure mode to watch for. Copy the structure and swap the subject, rather than copying the words. All of these run on the free 50 credits a day.
Flowain · Last reviewed 2026-09-03
What separates a prompt that works from one that does not
Every prompt below follows the same skeleton. Learn the skeleton and you can write your own.
The structure is: shot size, subject, action, setting, light, camera. Six slots. A prompt that fills all six gives the model very little to guess at. A prompt that fills two leaves the other four to chance, and you pay a full generation to find out what it chose.
Notice what is absent from every example here. There are no words like cinematic, beautiful, stunning, masterpiece, or 4K. Those are value judgements rather than instructions, and they consume prompt space without steering anything. Resolution is a generation setting, not a prompt term.
Each example names its mode, because mode choice matters as much as wording. Trying to force a Frames to Video shot through a text prompt is the most common reason people burn credits on retries.
Six shot setups, explained
Each one names the mode, the prompt, why it is built that way, and what usually goes wrong.
Slow push on a single subject
Text to videoMedium shot, a woman in a red wool coat stands still on a rain soaked street at night, neon shop signs reflecting in the puddles around her, slow dolly in, shallow depth of field, practical light sources only.
Why it is built this way. The subject is not moving, so the only motion is the camera. That is the easiest kind of shot for the model to keep coherent, which makes it a good first generation when you are learning how a prompt reads.
What to watch for. Neon reflections are where this one tends to break. If the puddle reflections drift independently of the signs, the shot will feel wrong even though nothing is obviously deformed.
Locked off product cutaway
Text to videoClose up, a ceramic coffee cup on a pale oak table beside a window, steam rising slowly, soft side light from camera left, locked off camera, deep focus, quiet room tone.
Why it is built this way. A locked off camera removes camera motion as a variable entirely, so the only thing being generated is the subject and the light. For product work this is usually what you want, and it fails less often than a moving shot.
What to watch for. Steam is the tell. It should rise and dissipate rather than loop. Native audio will add room tone here, which is often more convincing than the picture.
Frames to Video transition
Frames to VideoFirst frame: wide shot of an empty desert road at dawn. Last frame: same road, same angle, a single car now centred in the middle distance. Generate the approach.
Why it is built this way. This is the highest control setup in Flow. You have fixed both endpoints, so the model is only solving for the motion between them rather than inventing a composition. It is usually the cheapest route to a shot you already have in your head.
What to watch for. Keep both frames on the same camera position unless you actually want a move. Two frames from different angles asks the model to invent a camera move as well as the action, which is where it gets expensive.
Dialogue close up with native audio
Text to videoClose up, a man in his sixties sits in a parked car and says quietly, "I should have called sooner." Overcast daylight through the windscreen, handheld camera with slight drift, shallow depth of field.
Why it is built this way. Veo 3.1 generates dialogue alongside the picture, so the lip movement and the audio come from the same pass. Putting the line in quotation marks inside the prompt is what gets it spoken rather than described.
What to watch for. Short lines work better than long ones inside eight seconds. Anything past about a dozen words tends to get rushed or clipped at the end of the clip.
Same character across three separate shots
Ingredients to VideoSave the character first, then reference her by name in each shot: "@marta stands at the harbour wall at dawn, wide shot, slow pan right." Then "@marta unties a rope, close up, locked off." Then "@marta pushes the boat out, medium shot, handheld."
Why it is built this way. This is the shot type that shows what Flow is actually for. Three plain text generations of a woman at a harbour would give you three different women. Referencing a saved ingredient keeps it one person across all three, which is what makes them cut together.
What to watch for. Ingredients to Video runs at 8 seconds on Veo 3.1 Lite and Fast, and 4 to 10 seconds on Omni Flash. It is not available on Veo 3.1 Quality, so if the option is missing that is usually why.
Handheld tracking shot through a doorway
Text to videoMedium shot from behind, a figure in a canvas jacket walks away from camera through a doorway into a sunlit courtyard, handheld camera following, exposure shifting as they move into the light, warm bounce off stone walls.
Why it is built this way. Shooting a character from behind sidesteps the hardest thing for the model to hold steady, which is a face in motion. It is a practical trick for getting a usable character shot without spending generations on features that keep shifting.
What to watch for. The exposure change moving from interior to exterior is the interesting part and the part most likely to look artificial. Watch whether the shift happens at the doorway or arbitrarily.
How to adapt these for your own work
Swap the subject, keep the structure. That is the whole method.
Change one slot at a time. If you swap the subject, the setting, and the camera move all at once and the result is wrong, you have no idea which change caused it, and diagnosing that costs another generation.
Keep the shot size and camera move from an example that worked, and change only the subject and setting. Those two slots carry most of the creative difference and least of the technical risk.
When a scene needs more than one shot, do not describe the character again. Save them as an ingredient and reference them with @name in every prompt. Example five is built specifically to demonstrate that pattern, and it is the single habit that most improves the result.
Build your own library as you go. Veo responds consistently enough to phrasing that a note of what worked will save you real credits within a month.
Prompt patterns that hold up
Structural habits that reduce retries across almost any subject.
One camera move per generation
If your prompt contains the word “then”, split it. Eight seconds holds one move, and asking for two produces a shot that does neither well.
Name the light source
“Practical light from a desk lamp camera left” beats “moody lighting”. Direction and source are instructions; mood is not.
Shoot difficult subjects from behind
Faces in motion are the least stable thing Veo generates. A shot from behind or in silhouette sidesteps the problem entirely.
Keep dialogue short
A dozen words or fewer fits comfortably in eight seconds. Longer lines get rushed or cut off at the clip boundary.
Avoid readable text in frame
Signage and labels remain unreliable. Add any text you need in your editor afterwards rather than regenerating in hope.
Explore cheap, finish expensive
Find the shot on Lite at 10 credits, or Omni Flash from 4, then spend Fast at 20 or Quality at 100 only once you know it is the shot you want.
The step by step workflow these fit into is on the how to use Flow AI page, and the capability detail is on Flow AI features. If you have not used Flow before, what Flow AI is explains the tool and what the free tier covers before you spend a credit on it. If you are fitting these shots into a previsualization or client pipeline, Flow AI for designers covers where it sits alongside the tools you already use.
Flowain
Writer and editor, FlowAIFX
Flowain writes and maintains FlowAIFX
Every page is checked against Google's published specifications for Flow and Veo 3.1 on the date shown, and corrected when Google changes them. Where a limit is undocumented or a credit cost is easy to misjudge, it is written down rather than left out.
Last reviewed 2026-09-03 · Report a correction
Pick one and generate it
Start with the locked off product cutaway. It has the fewest moving variables, so it is the fastest way to see how your prompt wording maps onto what comes back.