- Why most Veo prompts come out looking flat
- The structure that gets you cinematic results
- A worked example
- Camera movement words that do something
- Lighting language that separates amateur from professional output
- Where this gets used in actual marketing work
- The mistake that wastes the most credits
- A short checklist before you hit generate
- Frequently asked questions
The short version: cinematic Veo prompts work because they specify camera, lens, light and movement in that order, not because they use fancy adjectives like "epic" or "stunning". Drop the mood words, add the technical ones, and your output changes dramatically. I'll show you the exact structure and a worked example below.
Why most Veo prompts come out looking flat
I've watched a lot of people type something like "cinematic shot of a woman walking through a city at night, moody, dramatic" into Veo and then wonder why it comes back looking like a stock video from 2015. The word "cinematic" does almost nothing on its own. It's become a filler word, like "nice" in a performance review. Veo, like most video models, responds far better to concrete technical instruction than to vibe words.
Here's the uncomfortable bit nobody wants to say out loud: the models are not reading your intent, they're pattern-matching your words against training data. If you say "moody" and nothing else, you get whatever the average moody clip in the dataset looks like, which is usually blue-toned and a bit generic. If you say "35mm lens, shallow depth of field, tungsten light bleeding from shopfronts, slow dolly push," you get something that looks like a director chose it.
The structure that gets you cinematic results
I build Veo prompts in a fixed order, and I'd suggest you do the same rather than freestyling a paragraph of adjectives. The order is:
- Shot type and camera movement (wide, close-up, dolly, pan, static)
- Lens and depth of field (35mm, 85mm, shallow focus, deep focus)
- Subject and specific action (not "a man," but "a man in his sixties tying his shoelace")
- Lighting source and quality (golden hour side light, fluorescent overhead, single key light)
- Setting detail (specific, not generic: "a closed launderette at 11pm" beats "a street")
- Colour and film stock reference if relevant (Kodak Portra tones, desaturated, high contrast)
This is the same logic I cover in more depth in my piece on camera prompts for AI, and it holds whether you're generating stills or moving footage. The camera and lens language is doing most of the heavy lifting.
A worked example
Say you run a small coffee roastery and you want a 6-second brand clip for Instagram showing the roasting process, something that feels like a short film rather than a product demo. A weak prompt looks like this:
"Cinematic shot of coffee beans roasting, warm and inviting, high quality."
That will probably give you beans in a vague machine with warm orange lighting, generic and forgettable. Here's the rebuilt version:
"Static close-up shot, 50mm lens, shallow depth of field. Coffee beans tumbling in a vintage drum roaster, steam catching the light. Single warm tungsten bulb overhead, casting long shadows across a concrete floor. Slow rack focus from the beans to a hand adjusting the roaster's temperature dial. Muted colour grade, slightly desaturated, like 16mm film stock."
Notice what changed: no mood adjectives left to do the work, every word is an instruction the model can execute. The rack focus and the specific light source give Veo something to animate deliberately rather than guess at.
A second quick example, for people-facing shots
For something with a person in frame, specificity about the person matters as much as the camera. "A young professional on a laptop" gives you the same three stock faces every model defaults to. "A woman in her forties with greying hair pulled back, reading glasses pushed up on her head, leaning forward slightly" gives you something far less generic. This overlaps with what I've written about in prompts for realistic people, and the same rule applies across video models, not just one tool.
Camera movement words that do something
Not all camera terms are equal in how reliably they render. From testing prompts repeatedly, these tend to be the most dependable:
- Dolly in/out - reliable, smooth, reads as intentional
- Slow pan - works well for establishing shots, less well for fast action
- Static shot - underrated, often looks more "cinematic" than movement because it forces composition to do the work
- Handheld - gives a documentary feel, useful for anything that needs to feel real rather than polished
- Crane shot - hit or miss, works best in wide outdoor scenes
If you only remember one thing from this list, remember that a static shot with good lighting usually outperforms a complicated camera move with vague lighting. People conflate movement with production value, and it's not the same thing.
Lighting language that separates amateur from professional output
This is where I think most guides on this topic are weak. They'll tell you to add "cinematic lighting" and leave it there. That phrase means nothing to the model. What works is naming the source and the quality:
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- "Single hard key light from camera left" (dramatic, high contrast)
- "Overcast daylight through a north-facing window" (soft, even, flattering)
- "Practical lamp light only, warm and dim" (intimate, low-key)
- "Backlit silhouette against a bright doorway" (strong mood, less facial detail)
Each of these tells Veo exactly where the light is coming from and what it's doing, rather than leaving the model to invent a generic three-point setup that looks like a corporate video.
Where this gets used in actual marketing work
If you're producing short clips for ads rather than brand mood pieces, the brief changes slightly, because you need a hook in the first second and clear product visibility, not just atmosphere. I've gone into that trade-off in my post on prompts for social media ads, but the short version is: cinematic and scroll-stopping aren't always the same goal, and you need to decide which one you're optimising for before you write the prompt.
For still images that need the same cinematic treatment, the camera and lighting logic transfers directly, and it's worth looking at how it applies to product photo prompts or even Instagram background prompts if you're building out a consistent visual set across stills and video.
The mistake that wastes the most credits
Veo generation isn't free and it isn't instant, so the biggest practical mistake is generating ten variations of a vague prompt instead of refining one specific prompt three or four times. I'd rather spend the first ten minutes writing the prompt, nailing shot type, lens, light and setting, than spend an hour regenerating guesses. Treat the prompt like a shot list a cinematographer would hand to a crew, not like a caption.
If you're doing this regularly for a business and it's eating hours every week, that's usually the point where it makes more sense to bring in someone who already has the prompt libraries and workflow built, rather than reinventing it each time. That's exactly the kind of gap an AI consultant for small business work tends to close quickly.
A short checklist before you hit generate
- Have you named a shot type, not just "cinematic"?
- Have you named a lens or depth of field?
- Have you described the light source, not just the mood?
- Is your setting specific enough that two people would picture the same place?
- Have you removed vague adjectives that aren't doing any instructional work?
If you can answer yes to all five, you'll get something closer to a director's shot than a stock clip almost every time.
Frequently asked questions
What's the single biggest mistake people make with Veo prompts?
Relying on mood words like "cinematic," "epic" or "stunning" instead of technical detail about camera, lens and light source, which is what the model responds to.
Do longer Veo prompts produce better results?
Not automatically. A tightly written prompt with five or six specific, technical details beats a long paragraph of vague description every time.
Should I specify camera movement in every Veo prompt?
No. A well-lit static shot often looks more professional than a camera move with weak lighting, so only add movement when it serves the shot.
How is writing cinematic prompts for Veo different from Midjourney or Sora?
The core logic of naming shot, lens and light is the same across tools, but Veo also needs clear instruction on movement and timing since it's generating motion, not a single frame.
Run a business, a brand or a blog? I feature tools, products and expert guest posts for an audience of 15,000. See how it works