How Do You Write a Prompt for an AI Video?

A working AI-video prompt is a single sentence with five parts in a fixed order: shot scale, subject, action, setting and light, and atmosphere at the end. In Yeşilçam Studios the camera move, aspect ratio, duration, and quality tier are chosen with buttons, so they stay out of the text.

A working AI-video prompt is a single sentence with five parts arranged in a fixed order. Shot scale, subject, action, setting and light, and — at the very end — atmosphere. In Yeşilçam Studios the camera move, aspect ratio, duration, and quality tier are chosen with a button, so they stay out of the text. Everything else has to be written.

The example below was written by the studio itself. On 2 September 2026, Flow Studio turned a 62-character Turkish synopsis — an elderly ticket conductor boarding the last tram on a rainy evening — into this 268-character shot prompt.

Medium shot, an elderly tram conductor in a dark vintage uniform stepping onto an old Istanbul tram at night, rain dripping from the roof, wet cobblestones reflecting warm street lamps, cinematic moody atmosphere, shallow depth of field, slight handheld camera motion.

That sentence is the template. The five parts sit inside it in order.

Why does the order matter?

A five-second clip has room for one idea, and the studio's own prompt shows the order that idea should be written in. Shot scale comes first, because how much of the frame the subject fills is set by camera distance and lens, and everything else in the image is placed around it.

Shot scale. Medium, close-up, or wide. One term, right at the front.

Subject. A single person or object with two or three visible details. "A dark vintage uniform" says more than "a man."

Action. One verb in the present tense. Stepping onto the tram. Two actions in one sentence compete for the same five seconds.

Setting and light. Where, when, and what the light is doing. Rain dripping from the roof, wet cobblestones, warm street lamps.

Atmosphere. Two or three words at the very end. Cinematic, moody, shallow depth of field.

What does the studio add on its own?

Four things are set outside the text. Writing them into the sentence means setting up a second time a control a button already handles.

Camera move. Thirteen options from Free to handheld, one per clip, chosen in the Create panel. The example sentence ends with a handheld move because that's how Flow Studio wrote it; in Create the same request is a button, and the camera-moves guide says which one to press.

Aspect ratio, 1:1, 16:9, or 9:16.

Duration and quality tier, which together set the price. A five-second draft clip is 490,000 tokens at the base rate; a four-second Cinema Pro clip is 1,872,500 — whatever the prompt says.

Speech. The "speech ready shot" switch adds a forward-facing face, a closed mouth, and a slight head motion to the prompt on its own, so the character can be lip-synced later.

Characters are the fifth thing. They come from the Characters tab and are attached to the shot rather than described in the prompt. Describing a recurring face in words is the answer to why two shots of the same person rarely match, and the character-consistency guide explains it in detail.

How long should the prompt be?

The studio's own example is 268 characters, about 40 words. The working range is 30 to 60 words. Flow Studio's script field takes 8,000 characters, but that's an entire film — the studio splits it into scenes and writes one prompt per scene. That's where long text belongs, never a single shot.

Words describing quality do nothing. 4K, ultra-realistic, and high-resolution are each a setting. The tier sets the engine, and the render size is chosen separately with 400, 800, or 2,400 tokens for 720p, 1080p, and 2160p. Every word spent on them is a word not spent on the subject.

Where the prompt stops

Here's the limit: the Create panel has no negative-prompt field, no seed, and no style-preset menu, so the look has to be written into the sentence every time, and a look that finally works has to be carried by hand. That's exactly why the Library keeps a Copy prompt button on every image and clip.

A prompt is also more a request than an instruction. A shot you throw away burns the same money as the shot you keep — 490,000 tokens at the cheapest video tier. A 29,750-token draft image tests the same sentence at a sixteenth of the price, which is why the first three versions of any prompt belong in image mode.

FAQ

Can I write the prompt in Turkish? Flow Studio accepted a Turkish synopsis and wrote the shot prompt in English, and that prompt stays editable before the scene is generated. The Create panel's placeholder text is English, and English is the safer option for a single shot.

Does a longer prompt spend more tokens? No. Price is set only by quality tier and duration. A ten-word prompt and a sixty-word prompt both cost 665,000 tokens for a five-second standard clip at the base rate.

How do I reuse a prompt that works? Every item in the Library has a Copy prompt button. Paste it into the Scene field, change one of the five parts, and keep the other four.

Related in Lodos

Data Analyst Web Analyzer Antivirus Lodos for Developers Lodos for Marketing Teams
30+Modules
4.9K+Users
FreeTo Start

Put it into practice.

Everything covered in this article is built into Lodos - one workspace, zero extra subscriptions.

Switching from another tool? Slack · Notion · Zoom · Jira · Postman · Toggl · Google Drive

More from the blog