How to Write AI Video Prompts: A Complete Guide with Templates and Examples

You typed something into an AI video generator, hit generate, and got... a slightly blurry person standing completely still. Sound familiar?

The tool isn't broken. The prompt is just missing the information AI actually needs to build a moving scene. Unlike image prompts, AI video prompts have to account for time — what starts the shot, what moves, how the camera behaves, and where everything lands at the end. Get that right, and the results improve dramatically.

This guide walks you through the full process: what to decide before you write a single word, a six-element formula you can reuse across any project, prompt examples for four common video types, and a troubleshooting section for when things still go sideways.

Jumping straight to the prompt field is tempting, but the three decisions below will save you multiple failed generations.

Define the Video Goal and Target Platform

Where the video lives determines its pacing, framing, and visual weight — a landing-page showcase and a 15-second Reel are completely different briefs. Before you write anything, ask: what do I want viewers to do after watching, and where will they watch it? That answer shapes every other choice in the prompt.

Choose the Right AI Video Generation Method

Most AI video tools — Runway, Kling, Sora, Pika, Veo — support three input modes, and each one changes what your prompt needs to do.

Text-to-video builds everything from scratch based on your words alone. You're the only source of visual information, so your prompt needs to cover the full picture: subject appearance, environment details, action, camera work, and style.

Image-to-video starts from a photo or illustration you upload. The visual foundation already exists, so your prompt shifts focus to motion: what moves, how it moves, and where the camera goes. Redescribing what's already visible in the image usually causes more confusion than it solves.

First-and-last-frame generation lets you provide both an opening and a closing image, with the AI filling in the transition. Here, your prompt describes the logic of that middle section — the speed, the kind of movement, and how the two shots connect.

Pick your method first. It tells you exactly what the prompt needs to carry and what it can leave out.

Set the Aspect Ratio, Duration, and Output Requirements

These are easy to overlook and annoying to fix after the fact. Lock them in before you write:

• Aspect ratio: 16:9 for YouTube and web, 9:16 for Reels, TikTok, and Shorts, 1:1 for certain feed posts

• Duration: Most tools cap a single generation at 5–10 seconds. If you need longer, plan to generate in segments and edit them together.

• Resolution: Check whether your tool supports 1080p or 4K output and whether that affects credit cost or generation time.

Most of these get set in the tool's interface — but knowing them beforehand means you won't generate a gorgeous 16:9 clip for a vertical placement.

Building your prompt around a timeline — what opens the video, what moves, and where it ends — is what separates motion from a fancy still.

Describe the Opening Shot

The opening shot is your foundation. AI uses it to establish the visual logic for everything that follows, so be specific. Describe the subject's position and state, the environment around them, the lighting conditions, and the initial camera framing.

"A glass perfume bottle on a white marble surface, side angle medium shot, soft studio light from the left, blurred neutral background" gives the AI a clear visual anchor to build from. "A perfume bottle" does not.

Explain the Main Action and Movement

This is the part most prompts get wrong — or skip entirely. If you don't describe what moves, the AI often defaults to a nearly static shot with minimal motion — which usually isn't what anyone wanted.

Be direct and specific about what's happening: the subject walks left to right, steam rises from the cup, the camera pushes forward, the flower opens. Include pace — "slowly," "in a single fluid motion," "over the course of the shot" — because speed is part of the visual experience too.

Specify How the Camera Follows the Subject

Camera movement is a separate dimension from subject movement, and the two can work together or independently. Common camera directions worth knowing:

• Zoom in / push in — moves toward the subject, building focus or tension

• Zoom out / pull back — reveals more of the environment, creates scale

• Pan left / pan right — horizontal sweep, good for following motion or revealing space

• Tilt up / tilt down — vertical movement, great for revealing height or grounding a scene

• Tracking shot — camera follows the subject as it moves

• Orbit / arc — camera circles the subject, useful for product showcases and hero moments

If you don't specify, the AI will choose for you — and its default is often a static lock-off or a generic slow push that may not fit what you had in mind.

Define the Final Shot

Describing where you want the video to end helps the AI work backwards and plan the motion in between. It doesn't need to be elaborate — "ending on a close-up of the product label" or "final frame: wide shot with the subject centered against the skyline" gives enough direction.

Some tools support first-and-last-frame mode for this exact purpose, which handles the end state more reliably than describing it in text. If visual consistency at the end matters for your project, that mode is worth using.

Add Constraints to Maintain Visual Consistency

Telling the AI what NOT to do is just as useful as telling it what to do. This is where negative prompts come in — a list of elements you want to exclude or behaviors you want to prevent.

Common exclusions for quality: blurry, low quality, watermark, grainy, overexposed

Common exclusions for character integrity: extra fingers, distorted hands, deformed face, inconsistent clothing

Common exclusions for scene control: no additional people, no background text, no camera cuts

Most tools have a dedicated negative prompt field. Use it. It's one of the fastest ways to reduce unwanted surprises in your output.

Every solid AI video prompt covers six elements. Knowing what each one does lets you decide consciously what to include and what to trim.

The template: Subject + Action + Setting + Camera + Visual Style + Technical Controls

Subject: Who or What Appears in the Video

The subject is the main visual focus — a person, an object, an animal, a landscape. Describe it well enough that the AI can build a consistent image: physical appearance, size, color, position in the frame, and any distinctive details that matter.

The more the subject needs to stay consistent across the shot (especially for faces and hands), the more descriptive you need to be here. Vague subject descriptions lead to drift — the subject changes subtly as the camera angle shifts, which is distracting in a way that's hard to fix in post.

Action: What Happens and How the Subject Moves

Describe the visible motion: what the subject does, how fast or slow, and in which direction. If the background also moves (leaves in the wind, crowds flowing, water rippling), describe that too — it won't happen automatically.

Action language that tends to work well: "walks steadily toward the camera," "slowly tilts its head upward," "the liquid pours in a single unbroken stream," "steam drifts upward and disperses." Avoid internal states ("feels nervous") and abstract descriptions ("embodies energy") — AI reads prompts literally and won't know what those look like on screen.

Setting: Where and When the Scene Takes Place

Setting covers location, time of day, weather, and the environmental details that surround the subject. The more specific you are, the less the AI will fill in gaps on its own — which is usually what you want.

"A minimalist café" is a starting point. "A small Tokyo-style coffee bar with warm pendant lighting, exposed concrete walls, and a rain-fogged window in the background, late afternoon" is a scene. The second version gives the AI less room to interpret and more room to render.

Camera: Shot Size, Angle, and Camera Movement

Camera direction is one of the highest-leverage elements in any video prompt. The same subject and action feels completely different when shot from a low angle versus an overhead view, or with a slow push-in versus a static lock-off. Shot size runs from extreme close-up through establishing shot; angles from eye-level to bird's-eye; movement options are listed in the previous section.

If camera work matters to the feel of the video, be explicit — otherwise you're leaving it up to chance.

Visual Style: Lighting, Mood, Color, and Aesthetic

This is where the video gets its personality. Lighting descriptors like "golden hour warmth," "cold blue night," "harsh overhead fluorescent," or "soft diffused studio light" tell the AI how to sculpt the scene emotionally. Color palette notes like "desaturated earth tones" or "high-contrast neon" shape the mood fast.

For overall aesthetic, describing the feel tends to work better than naming specific films or directors — "cinematic with shallow depth of field," "documentary-style handheld," "clean commercial aesthetic," or "dreamy soft-focus" all translate more reliably than "make it look like a Villeneuve film."

Technical Controls: Duration, Sound, Effects, and Constraints

This element covers everything that's more mechanical than visual. If your tool supports it, you can specify:

• Duration: "5-second clip" or "generate approximately 8 seconds"

• Negative prompts: listed in a dedicated field or added with phrases like "avoid..." and "no..."

• Effects: motion blur on fast movement, lens flare, film grain, depth of field

• Sound or audio cues: some newer tools (like Kling 3.0 with Native Audio) accept audio direction alongside video

These details clean up the result and keep it from drifting into territory you don't want.

The same video idea needs a differently weighted prompt depending on how you're generating it. Here's how the focus shifts across the three main modes.

Text-to-Video Prompts: Describe the Complete Scene

Text-to-video means the AI has zero visual context — cover all six elements with enough specificity that it doesn't have to guess. That doesn't mean writing an essay. A well-structured 80-word prompt usually outperforms a vague 200-word one. Prioritize the most visually critical details first: subject, action, and camera tend to carry the most weight in how the output turns out.

One practical tip: place your most important descriptors early in the prompt. Most models weight the beginning more heavily than the end.

Image-to-Video Prompts: Focus on Motion and Camera Movement

With a reference image, subject and setting are already defined — your prompt's only job is motion and camera. Re-describing what's already visible ("the woman is wearing a blue dress, there's a wooden table...") often confuses the model rather than helping it. Skip the visual inventory and go straight to motion: "the steam rises slowly from the cup, camera gently pushes in, holding focus on the surface of the coffee."

If there's one change to make and one thing to leave alone, state both clearly: "the candle flame flickers gently, everything else remains still."

First-and-Last-Frame Prompts: Describe the Transition Between Images

Your prompt should describe the transition logic: what kind of movement connects the two images, how fast it happens, and whether the path is smooth or dynamic.

One important thing to check before generating: the two images should share enough visual continuity (lighting direction, color temperature, environment scale) that a physical transition between them is plausible. If the starting frame is a close-up in warm light and the ending frame is a wide shot in cool blue tones, the AI will struggle to bridge them naturally no matter how good the prompt is. In that case, describe the transition logic explicitly — "a slow fade combined with a gradual camera pull-back" can help paper over larger visual gaps.

Here are four common video types, each with a before-and-after prompt.

Product Showcase Video Prompt Example

Vague prompt: A skincare serum ad video.

Improved AI video prompt: "A small glass serum bottle sits on a white marble surface, shot from a 45-degree angle at medium distance. A single droplet forms at the dropper tip and falls in slow motion into still water below, creating a soft ripple. The camera slowly pushes in toward the droplet impact. Lighting is soft and diffused, coming from the upper left, with a subtle warm glow. Background is clean white with a slight gradient shadow. Style: premium beauty commercial, shallow depth of field. No hands, no text in frame."

Character Action Video Prompt Example

Vague prompt: A man walking through a city at night.

Improved AI video prompt: "A man in his late 30s wearing a dark overcoat walks along a wet city sidewalk at night, moving steadily from left to right across the frame. The camera tracks alongside him at mid-shot height, keeping pace. Neon signs and streetlights reflect on the wet pavement. The background shows shallow-focus pedestrians and passing car lights. Mood: quiet and cinematic, slightly melancholic. Color palette: deep blues and amber. No sudden camera cuts."

Food Commercial Video Prompt Example

Vague prompt: A pasta video for Instagram.

Improved AI video prompt: "A bowl of freshly plated carbonara sits on a dark slate surface. Steam rises slowly from the noodles as a hand twirls a forkful in a slow, deliberate rotation — the camera follows the fork from medium shot down to a tight close-up as it lifts from the bowl. Warm side lighting from the right creates a golden sheen on the pasta surface. Background is softly blurred kitchen environment. Mood: indulgent and tactile. Style: high-end food editorial, warm color grading."

Social Media Video Prompt Example

Vague prompt: A lifestyle video for Reels.

Improved AI video prompt: "Opening on an extreme close-up of a ceramic mug being set on a wooden table — the sound of a soft clunk implied. Camera pulls back slowly to reveal a person sitting by a large rain-streaked window in the early morning, both hands wrapped around the mug. City rooftops visible through the glass. Handheld camera feel with minimal shake. Color palette: cool blues and soft warm amber from the mug. Duration: approximately 6 seconds, 9:16 vertical framing. Style: quiet lifestyle, documentary-adjacent."

Even a well-built prompt won't always land on the first try. Here's what's usually going wrong — and how to fix it.

The Video Has Little or No Movement

This is the most common failure mode. The AI generated a beautiful static image that barely moves.

Why it happens: No explicit action language — without clear motion cues, most models default to minimal movement.

Fix: Add specific action verbs with pace descriptions. "Steam rises slowly," "the character walks toward the camera," "the camera pushes in over 5 seconds" — these give the model something to animate. If the tool has a motion intensity slider or equivalent setting, increase it alongside the prompt change.

The Character's Movements Look Distorted

Arms bend at odd angles, fingers multiply, faces blur mid-motion — character movement is still one of the harder things for AI video models to get right.

Why it happens: Complex or fast movements, multi-person interactions, and detailed hand actions push current models to their limits.

Fix: Simplify the action. Instead of "the chef chops vegetables rapidly," try "the chef stands at a counter, making a slow deliberate movement with both hands." Keep the subject's action to a single direction or motion per generation. Avoid close-ups of hands doing anything intricate unless the tool specifically handles it well.

The Camera Movement Is Too Strong

You asked for a gentle push-in and got a whiplash zoom.

Why it happens: Words like "push in" or "zoom" carry no default intensity — the model picks the magnitude on its own.

Fix: Add pace qualifiers: "very slowly," "subtly," "a slight drift," or "over the full duration of the clip." You can also anchor the camera by specifying start and end framing: "beginning as a medium shot, ending just inside close-up range" is more precise than "push in."

The Scene Changes Unexpectedly

The video starts in one location and cuts to somewhere completely different.

Why it happens: Multiple location or environment descriptions in a single prompt get interpreted as scene changes rather than one continuous scene.

Fix: Keep each generation to a single setting. If your video concept requires multiple locations, generate each separately and edit them together in post. Also avoid phrases like "then," "next," or "followed by" in text-to-video prompts — they signal a sequence to the model, which may read as a cut cue.

The Subject's Appearance Is Inconsistent

The character looks noticeably different between the beginning and end of the clip — different hair, face, or clothing.

Why it happens: Models struggle to maintain visual continuity on the subject as camera angle shifts or the clip runs long.

Fix: Keep the motion arc simple — subjects that move primarily in one direction (rather than turning to face a new angle) stay more consistent. In the prompt, emphasize the most distinctive visual traits: "maintaining the subject's red coat and short dark hair throughout." If the tool supports image or video element references, use them — they anchor the subject's appearance far more reliably than description alone.

The Video Does Not End on the Specified Shot

The final frame is nowhere near what you described.

Why it happens: Most text-to-video models treat end-state descriptions as soft suggestions, not firm constraints.

Fix: Switch to first-and-last-frame mode if your tool supports it — this is the most reliable way to control the ending. If you're working in text-to-video mode, try placing the end-state description near the beginning of the prompt rather than the end: models tend to weight early content more heavily, which occasionally improves end-frame accuracy. Alternatively, generate a second clip that starts where you want the first one to end, then cut them together.

Writing a full structured prompt from scratch takes practice — and when you're moving fast or working across multiple video formats, the setup time adds up. GenApe's AI video prompt generator is built to close that gap.

Describe your concept in plain terms, and GenApe builds out a complete, structured prompt optimized for the AI video tool you're using — covering subject, action, camera, style, and technical controls without requiring you to know every term or format by heart. It also offers prompt templates by video type, so whether you're generating a product demo, a social media clip, or a cinematic scene, you're starting from a structure that works instead of a blank field.

If you're ready to stop guessing and start generating, GenApe's prompt assistant is the fastest way to go from idea to output.