> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://help.seelab.ai/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Prompting a video: intact product, camera and lenses

In video, the prompt doesn't describe an image: it directs a shot. What moves, what stays still, where the camera goes, with which lens. This guide gathers the rules that work across the engines of the **Video** space for successful [product videos](https://www.seelab.ai/en/use-cases/product-video), illustrated with a speaker from the demo brand Nord Audio.

${youtube}[Prompting a video](yHE0hWRjF4I)

# The structure of a good video prompt

When you start from an image, it already contains the product, the setting, the light and the framing. The prompt doesn't need to describe them again: it describes **what changes** during the video.

Build it in this order:

1. **The subject and its action**: one main action per shot.
2. **The camera**: one single move, with its speed ("slow", "steady").
3. **The scene**: what moves around it (reflections, smoke, leaves, passers-by).
4. **The style**: the lens, the light, the grain.
5. **What must stay identical**: the product, the logo, the materials.

> Product shot on a studio background. The speaker stays perfectly still on the slate base. The camera makes a slow 30-degree arc to the right, at product height, in a smooth, steady move. A highlight glides slowly across the aluminum top. The NORD AUDIO logo, the fabric weave and the cork base stay identical from start to finish.

In the **Medias** tab of the Video space: the start frame (1), the engine (2), the prompt (3), the camera presets (4), then **Generate** (5).

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-ui-medias_1k7jie.jpg =1000xauto)

| **Phrase it positively.** Engines struggle with negations: "the camera stays locked" works better than "no camera movement".

A few more reflexes:

* **One shot, one idea.** Too many actions in 5 seconds make a confusing video. To chain several actions, increase the duration and give time cues: "in the last two seconds…".
* **With a start and an end frame**, describe the transition between them, not the frames themselves. Keep the two frames close: to go from day to night, make two videos instead.
* **Change one thing at a time** when you fix a prompt: you'll know what improved the result.

# Keep a product intact, down to the detail

It's the number one request from brands: a video where the bottle, the label, the logo and the material don't move an inch. The golden rule: **move the camera and the light, not the product.**

Same start frame, same engine (Kling v4), two prompts:

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-compare_1xa7sj0.jpg =1000xauto)

**At the top**, the prompt asks the product to spin, fly up and be flipped over:

> The speaker spins quickly on itself, rises into the air, then a hand grabs it and flips it over to show the bottom, dynamic lighting effects.

The engine has to invent sides it has never seen: the logo gets lost, the proportions change.

**At the bottom**, the product stays still and the camera moves around it (prompt above, in "The structure of a good video prompt").

The rules that protect your product:

* **Start from an approved image**, sharp, with the product fully visible. It is the most reliable method.
* **Say the product is still**: "the speaker stays perfectly still on the base".
* **Get rotation from the camera**: a slow 20 to 40-degree arc rather than a product spinning on itself.
* **Avoid revealing hidden sides**: a full turn forces the engine to invent the back of the product.
* **Name what must stay identical**: the logo, the label text, the materials, the colors.
* **Don't describe the product again**: a description that differs from the image creates hesitation, and the product drifts.
* **Launch 2 to 4 videos** and keep the one where the label is perfect.

| **Even more reliable: your Seelab model.** In the **Custom Models** tab, the video version of your [product model](https://www.seelab.ai/en/platform/product-modeling) already knows the shape, logo and colors of your product. You only describe the scene and the motion.

# Direct the camera

One move per shot, described with film-set vocabulary, plus a speed. Engines understand these terms very well:

| Move | What it does | What to write |
| --- | --- | --- |
| Static shot | Nothing moves except the subject | "locked-off camera, static frame" |
| Push in / pull out | The camera moves forward or back | "slow push in toward the logo" |
| Truck | The camera slides sideways | "truck from left to right" |
| Arc, orbit | The camera circles the subject | "slow 30-degree arc to the right" |
| Crane | The camera rises or lowers | "the camera rises and ends top-down" |
| Pan | The camera pivots in place | "slow pan to the right" |
| Handheld | Lively, slightly unsteady motion | "handheld camera, documentary shot" |
| Drone | Flying overhead | "drone shot moving forward above…" |

Three moves from the same start frame:

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-camera_upr9x1.jpg =1000xauto)

> Product shot on a studio background, the speaker stays still on the base. Slow, continuous push in at product height, ending on a close-up of the NORD AUDIO logo.

> The camera rises slowly and vertically, like on a crane, and ends top-down above the product.

> Slow truck from left to right, the camera glides parallel to the product at a constant distance.

* **A zoom is not a push in**: a zoom changes the framing without moving the camera, a push in creates depth.
* **Give the end point**: "…ending on a close-up of the logo".
* **No time to write?** The **Camera Presets** button, under the prompt, adds a predefined move in one click.

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/05-03camera-presets_591dms.jpg =500xauto)

# Pick the lens

The lens changes the whole feel of a shot. Name it like a cinematographer: focal length, aperture, effect.

| Lens | The effect | Best for |
| --- | --- | --- |
| Macro 100 mm | Details a few centimeters away, very blurred background | Textures, materials, finishes |
| Wide 24 mm | Strong perspective, very present setting | Product in its environment |
| 50 mm | Natural view, close to the human eye | Neutral shots, e-commerce |
| 85 mm f/1.4 | Soft bokeh background, subject stands out | Portraits, hero product |
| Anamorphic | Horizontal flares, oval bokeh, cinematic look | Brand films, night moods |

Four lenses, same product, with the speaker's Seelab video model:

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-lenses_d57as9.jpg =1000xauto)

> Macro shot, 100 mm macro lens. The camera glides very slowly along the speaker's fabric: you can see each thread of the weave, then the brushed edge of the aluminum. Very shallow depth of field, raking studio light, dark background.

> Anamorphic lens, cinematic look. The speaker stands on the counter of a cocktail bar at night. Blue horizontal flares from the neon lights, oval bokeh in the background, film grain. Slow truck from left to right.

* **Write "85 mm lens, f/1.4"** rather than "photorealistic": it's more precise and more stable.
* **Pair lens and motion**: macro + very slow glide, wide angle + push in, anamorphic + truck.
* **Starting from a start frame?** It already sets the lens. To change lenses, use the Seelab model, Multi Refs or the Edit tab.

# Light and mood

Name the light concretely rather than with adjectives: "low morning light from the left", "backlight at sunset", "blue and pink neon", "soft, diffused studio light". Keep the same light and grain words from one video to the next: the whole campaign will share the same signature.

# Edit rather than regenerate

A video that's right except for one detail? Don't relaunch it: the **Edit** tab keeps the motion and the framing, and only changes what you ask.

1. **Open the Edit tab**: the pencil on a video in the feed sends it there as the source video (1). Add an image in Multi Refs if needed (2) and choose the engine, here Kling v4 (3).
2. **Write the change** (4), then launch.

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-ui-edit_riii3h.jpg =1000xauto)

> Same shot, same camera movement. Replace the studio background with a bright Scandinavian living room at sunset, the speaker standing on a light oak coffee table, low golden light through the window. The speaker, its logo and its materials stay identical.

![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/vm-en-edit_1w8co51.jpg =1000xauto)

* **Start with what changes, then what stays**: "Replace the background with… The speaker stays identical".
* **One change per pass**: setting, then light, then color.
* **To add a specific element**, put it in **Multi Refs** and name it: "the lamp from Image 1".

# Which engine for which shot

| Your need | The engine | The prompt tip |
| --- | --- | --- |
| Perfectly faithful product | Your Seelab video model | Describe the scene, never the product |
| Product that stays identical, up to 4K | SeeDance v2.0 | Start from an approved image, slow camera |
| Long shots, materials, human motion | Kling v4 | One move per shot, time cues beyond 10 s |
| Approved visual to animate with sound | Gemini Omni Flash 1.2 | 3 or 4 clear sentences rather than a list of adjectives |
| Multi-shot film with dialogue | SeeDance v2.5 | One action per shot, numbered shots |
| Transition between two images | Minimax H3-Max | Describe the passage from image 1 to image 2 |

All engine sheets are in [Video models in detail](https://help.seelab.ai/en/article/video-models-in-detail-ie0uck/).

# Cheat sheet

* Describe the **motion**, not the image.
* **One subject, one action, one camera move** per shot.
* The product is **still**, the **camera** and the **light** move.
* Name **what stays identical**.
* Talk like a cinematographer: **focal length, aperture, move, speed**.
* Phrase it **positively**.
* An almost perfect video gets **edited**, not remade.

# Learn more

* [Generate a video](https://help.seelab.ai/en/article/how-to-generate-video-with-seelab-15wg3ir/)
* [Animate your product with your video model](https://help.seelab.ai/en/article/animate-your-product-with-your-video-model-1f7ismq/)
* [Edit a video](https://help.seelab.ai/en/article/edit-a-video-with-example-prompts-7fi5gh/)
* [Which video model to choose?](https://help.seelab.ai/en/article/which-video-model-to-choose-kim2xb/)