18 second product ad
A complete walkthrough of a vertical product ad where the product looks the same in every shot, with captions and platform exports.
This recipe makes an 18 second vertical ad for a physical product. The product must look identical in every shot, so every shot starts from a real photo of it. Kling 3.0 on the Standard plan covers the whole ad with credits to spare. Seedance 2.5 is the alternative if you prefer one multi shot generation.
The brief
| Item | Choice |
|---|---|
| Product | NAMI, a 750 ml insulated steel bottle in matte sage green with a bamboo cap (a made up example; use your own product) |
| Message | Your water stays ice cold in a hot car |
| Length and format | 18 seconds, 9:16, for TikTok, Reels and Shorts |
| Structure | Hook in the first 2 to 3 seconds, problem, product moment, payoff with a call to action |
| People | None. No hands, no faces. This removes the most common failures. |
| Text | All text and the logo added in the edit, never generated |
Budget
| Item | Kling 3.0 (Standard plan) | Seedance 2.5 (API rates) |
|---|---|---|
| Product photos | Your own photos, or one packshot from an image tool | Same |
| Drafts | 4 shots x 3 takes x 5 s at 720p, silent: 360 credits | 3 takes of the full 18 s at 480p: about 5.60 USD |
| Finals | 4 shots x 5 s at 1080p, silent: 160 credits | 1 take of 18 s at 720p: about 4.20 USD |
| Total | 520 of the 660 monthly credits, about 5.50 USD in the first month | About 9.80 USD |
Prices checked on 15 Oct 2026. Kling 3.0 costs 6 credits per second at 720p and 8 at 1080p without native audio. Kling Standard is 6.99 USD for the first month, then 8.80 USD, for 660 credits, with 1080p, commercial use and no watermark. The free Kling tier is watermarked and not for commercial use, so make the finals on a paid plan. Seedance figures use the official ModelArk rates (0.103 USD/second at 480p, 0.231 at 720p) as a yardstick; Dreamina shows its own credit cost before you generate.
Tool stack: a phone camera or an image tool for the packshot, Kling 3.0 (or Seedance 2.5 in Dreamina), CapCut for captions and export, a paid music source.
Walkthrough
Idea: problem first, product second
Lead with the viewer's problem, not the product. The hook is a cheap plastic bottle going soft on a hot dashboard. The product only appears after that, so the first 3 seconds earn attention without a logo.
Shot list
| # | Time in the ad | Framing and camera | Action | Text added in the edit |
|---|---|---|---|---|
| 1 | 0 to 3 s | Close-up, static | A clear plastic bottle on a sunlit car dashboard slowly sags in the heat haze | "2 PM. Parked in the sun." |
| 2 | 3 to 8 s | Medium close-up, slow push in | NAMI stands on the same dashboard in hard sunlight | "Still ice inside." |
| 3 | 8 to 13 s | Macro, static | The bamboo cap lifts off, ice cubes and cold vapor inside | None |
| 4 | 13 to 18 s | Packshot on wet black slate, slow push in | Bottle stands still, droplets catch the rim light | Logo and "NAMI 750 ml" |
Generate every shot at 5 seconds. Shot 1 is cut to 3 seconds in the edit.
Reference images
The product must come from an image every time. A product or logo described only in words comes out as fiction. Photograph the real product against plain backgrounds, or generate one clean packshot from your photo and use it for every shot.
Packshot of the attached bottle exactly as shown (shape, cap, label layout),
centered on wet black slate, single hard rim light from behind right, soft
fill from front, shallow depth of field, droplets on the steel, 9:16,
negative space in the top third for a title added in post. No added text.Then make the first frames. For shot 2, place the bottle on a car dashboard in sunlight. For shot 3, make two frames that differ as little as possible: the cap closed, and the cap lifted with ice visible. The more the first and last frames differ, the more the model invents.
Video prompts
For image to video, describe only what moves. Restating the image weakens the motion. End every prompt with a rest state; on Kling this also helps against generations that hang at 99%.
Use image to video with the first frame for each shot, and start and end frame for shot 3. Turn native audio off. Kling responds to plain directional words and timing cues like "At the 3rd second".
Close-up, static camera. A clear plastic water bottle stands on a car
dashboard in hard afternoon sun, heat haze shimmering above the dashboard.
The bottle slowly sags and its sides crumple inward. At the 4th second the
bottle stops moving and stays still. Warm orange light, dust in the air.Medium close-up, slow push in. The green steel bottle stands still on the
sunlit dashboard. Hard sunlight slides slowly across its matte surface and
the heat haze moves behind it. The bottle and its cap do not change shape.
At the 4th second the camera stops and holds.Macro shot, static camera. The bamboo cap lifts slowly straight up out of
frame. Cold white vapor curls out of the opening and ice cubes inside shift
and settle. The vapor fades and the frame comes to rest.Slow push in. The green steel bottle stands still on wet black slate. Water
droplets on the steel catch the rim light and one droplet runs down the
side. The camera stops at the 4th second and holds on the bottle.Make 2 to 3 drafts per shot at 720p, choose the keepers, and render the finals at 1080p.
Editing
Cut on motion: start shot 2 as the push in begins, and cut to shot 4 while the vapor is still moving. Trim the soft first frames of each clip. Hold the packshot for at least the last 2 seconds so the logo can be read.
Audio and captions
Generate silent and build the sound yourself. Add sound effects (heat ticking, the cap, ice) from ElevenLabs on a paid plan; free plan sound effects need attribution. For music in an ad, Eleven Music has the cleanest rights according to one licensing guide, and Suno songs are usable commercially only if made on a paid plan. Either one can still trigger YouTube Content ID.
Add the text from the shot list as captions in CapCut. Use bold sans serif text at about 7 to 10% of the frame height (roughly 54 to 80 px on a 1080x1920 frame). Burn the captions in, because many viewers watch muted. Keep text out of the platform interface: on TikTok, avoid the top 240 px, the bottom 660 px and 120 px on each side. These figures come from a guide citing TikTok Ads Manager and other tools give slightly different numbers, so check in the app before you post.
Finishing and platform export
Export one master and check it on a phone.
| Platform | Settings |
|---|---|
| TikTok, Reels, Shorts | 1080x1920 (9:16), 30 fps, H.264 MP4 with AAC audio, about 10 to 15 Mbps, loudness about minus 14 LUFS |
| YouTube (official) | MP4, H.264, AAC at 48 kHz; 1080p at about 8 Mbps |
The vertical settings are third party consensus; no official spec was found. Turn on the AI label when posting if the shots look real: TikTok has an "AI-generated content" toggle, YouTube asks you to disclose realistic scenes that never happened, and Meta shows an "AI info" label. Do not post the identical file everywhere and expect each copy to count as original; Meta down-ranks duplicates.
What usually goes wrong
| Problem | Why | Fix |
|---|---|---|
| The bottle changes shape or color | No image anchors the product in that shot | Start every shot from a product image and add "the bottle does not change shape" |
| Garbled text or a fake logo on the product | Video models render text poorly | Generate without text and add the logo in the edit |
| Kling stuck at 99% | The model does not know how to end the motion | Add an end state ("then holds"), keep 5 seconds, try a lower resolution or off peak hours |
| Different results each run on Kling | Kling is less predictable between runs | Generate 3 drafts and pick, instead of rewriting after one bad take |
| Unwanted music | Native audio fills empty slots | Generate silent, add sound in the edit |
| Captions hidden behind buttons | Text inside platform interface zones | Keep text centered and check the safe zones in the app |
Use your own product photos
If you sell a real product, shoot it on your phone in good light. Image models redraw what they see, so a generated packshot can drift from the real product in small details such as the cap or the label layout.
Sources
- Kling credit costs and plans: Kling 3.0 credit guide, Kling 3.0 user guide
- Seedance syntax and logo note: Seedance 2.5 prompt guide, Seedance 2.0 prompt guide, ModelArk pricing
- Product images and first and last frames: heyuan110, dev.to Veemo
- Kling quirks: VEED, fal
- Music and SFX rights: ElevenLabs pricing, Suno pricing, licenseorg
- Captions and safe zones: openclip, upload-post
- Export and labels: YouTube encoding, YouTube disclosure, TikTok newsroom