How Businesses Use AI Video Models for Product Ads
Learn how to create AI video ads with Seedance 2.0. See real-world tests, pricing, video formats, generation times, and practical tips.
A product video used to need a studio, a camera operator and a few days of editing. AI video ads models now produce a short ad from a text description or from a product photo in a few minutes, for a few dollars. This article explains where businesses use them, what the leading model family can do, and what it produced when we gave it three real ad briefs. We used Seedance 2.0 API from ByteDance, called through AI/ML API, and measured the price, waiting time and format of every clip.
Where Businesses Use AI Video Ads
The use cases that fit current models share one trait. They are short, they show one product or one scene, and they are needed in many variations.
- Vertical ads for social feeds. A 6 to 15 second clip in 9:16 for Instagram Reels, TikTok or YouTube Shorts.
- Product hero shots. A slow camera move around a product on a clean surface for a landing page or a marketplace listing.
- Lifestyle scenes. A product in use in a kitchen, an office or a bathroom, with sound, for brands that cannot afford a location shoot.
- Variations for testing. The same product in several settings, formats and lengths, so a team can see which version converts before it commissions a full shoot.
- Animated product photos. An existing packshot turned into a short video, which keeps the product exactly as designed.
What Seedance 2.0 can do
Seedance 2.0 API was released by ByteDance on 12 April 2026. It takes text, images or video as input and returns a video, with native audio and lip sync. According to the documentation, a clip can be 4 to 15 seconds long, 480p to 4K, and in six aspect ratios from 21:9 to 9:16, including vertical 9:16 for social feeds. You can add up to 9 reference images, 3 reference videos and 3 audio files to guide the result. Billing is per second of output, from $0.09243 to $1.014 per second depending on resolution.
The same family has three more models on AI/ML API. We did not test them, and the table below uses their documentation and model pages.
| Model | API model ID | Duration | Resolution | What differs | Price |
| Seedance 2.0 | bytedance/seedance-2-0 | 4 to 15 s | 480p to 4K | Standard model, best voice adaptation | $0.09243 to $1.014 per second |
| Seedance 2.0 Fast | bytedance/seedance-2-0-fast | 4 to 15 s | 480p, 720p | Lower latency, uses reference audio as it is | $0.07436 to $0.16003 per second |
| Seedance 2.0 Mini | bytedance/dreamina-seedance-2-0-mini | 4 to 15 s | 480p, 720p | Smaller option in the 2.0 line | Not listed in the pages we checked |
| Seedance 2.5 | bytedance/seedance-2-5 | 4 to 30 s | 480p, 720p | Released 7 August 2026, up to 50 reference assets, 30 second clips | $0.065 per second at 480p |
Our test, three ad briefs
We wrote three briefs the way a small business would. Two feature invented brands, LUMO for a skin serum and EMBER for a coffee roaster, so we could check whether the model keeps printed text readable. We ran every brief at 720p through the AI/ML API video endpoint, which takes the same two steps for every model. You send a request and check its status until the video is ready. The tests ran on 7 October 2026.
Ad 1, a vertical social ad for a face serum
Prompt: Vertical social media ad for a face serum. A green glass dropper bottle with a white label that reads LUMO stands on pale marble. Water droplets fall around it in soft morning light, the camera slowly pushes in, and the dropper releases one golden drop. Premium, clean, minimal.
The model produced a 6 second clip in 720 by 1280 pixels at 24 frames per second. It has a clear sequence, with droplets around the bottle, a push in on the dropper, and a single golden drop that lands on the marble in the last frames. The label reads LUMO in the first two frames and the bottle keeps its shape. The clip also contains an audio track that we did not ask for, because audio generation is on by default. It cost $1.40 and took 193 seconds, which is about $0.23 per second.
Ad 2, a lifestyle scene with sound
Prompt: A cozy kitchen at sunrise. A hand pours hot water over freshly ground coffee in a pour over filter, steam rises, and a ceramic mug stands next to a kraft paper bag with the word EMBER printed on it. Warm light, gentle camera movement, the sound of pouring water and soft acoustic guitar.
This brief asks for a full scene with a hand, a kettle, steam and sound, in 16:9 for 8 seconds. The model kept the word EMBER readable on the bag through the whole move and showed the water, the coffee bed and the steam in a consistent warm light. The file is 1280 by 720 pixels and has a stereo audio track at a mean level of minus 21 dB, so the sound is present and not silent. We have not rated the sound quality. It cost $1.80 and took 297 seconds, about $0.23 per second.
Ad 3, an existing product photo turned into a video
Most businesses already have a product photo, and a text prompt cannot reproduce a real package. For this brief we made a product photo with Flux Pro v1.1, which cost $0.052, and gave it to Seedance 2.0 as an image with a short instruction.
Prompt: The camera slowly pushes in while soft light moves across the marble. A single water droplet runs down the bottle. The label stays sharp and readable.
The bottle in the video matches the photo in every frame, with the same black cap, the same white label and the word LUMO sharp throughout. That is the main reason to start from a photo, because the text and the design stay as they were. The result also shows a limit. The droplet we asked for slides down the outside of the bottle from the cap, which is not how a serum behaves, so a person has to review the physics before the clip goes live. The clip is 5 seconds long at 834 by 1112 pixels. It cost $1.98 and took 268 seconds, about $0.40 per second, which is more than the text only clips. We did not find the reason for the higher price in the documentation.
What did not work
We also sent the serum brief to Seedance 2.0 Fast. The task failed with the message that the output audio may be related to copyright restrictions. Audio is generated by default, and the service can reject a request when the generated sound is flagged. If you plan to add your own music, set generate_audio to false. We could not rerun the task, and we did not run Seedance 2.5, because the test balance ran out. We do not know whether a failed task is billed, so check your balance after a failure.
What a clip costs
Related walkthrough
OpenAI Sora 2 – The Future of AI Video Creation | Next-Gen AI for Creators
| Clip | Length | Cost | Cost per second | Waiting time |
| Serum ad, text to video, 9:16 | 6 s | $1.40 | $0.23 | 193 s |
| Coffee ad with sound, 16:9 | 8 s | $1.80 | $0.23 | 297 s |
| Product photo to video | 5 s | $1.98 | $0.40 | 268 s |
| Product photo with Flux Pro v1.1 | 1 image | $0.052 | Not applicable | 7 s |
Which Seedance model to start with
| Need | Model | Note |
| Final ad at the highest quality or 4K | Seedance 2.0 | Tested by us at 720p |
| Many cheap drafts to choose from | Seedance 2.0 Fast | Cheaper per second and 480p or 720p only, our one test run was rejected by the audio filter |
| A clip longer than 15 seconds | Seedance 2.5 | Up to 30 seconds, not tested by us |
| Animating a real product photo | Seedance 2.0 with image input | Tested by us, keeps the label and the design |
Before you publish an AI product ad
Before you publish an AI product ad, review the final video carefully. Effective AI video production requires checking product accuracy, visual consistency, and advertising claims in every frame. Review the rules of the platform where the ad will run, since many platforms ask advertisers to label AI-generated content. Read the terms of the model about commercial use. Keep the prompt, the model name and the date for each clip, so you can reproduce or explain it later.
How to try it
All the Seedance models and many other video, image and text models are listed in the AI/ML API model catalog. One key reaches all of them, billing is pay-as-you-go, and the same two request steps work for each. The Seedance 2.0 API page lists the current price and the parameters for the model we used.
Conclusion
AI video ads offer businesses a faster and more affordable way to create product promotions, social media creatives, and lifestyle videos. While AI video models can simplify production and make creative testing more accessible, every clip still needs human review for product accuracy, realistic motion, and brand consistency.
Frequently Asked Questions
For short social clips and variations it can, as our three tests show. For a hero campaign with real people and exact product handling, a shoot is still the safer choice, and a person must review every clip.
In our test a 6 second vertical ad cost $1.40, an 8 second ad with sound cost $1.80 and a 5 second animated product photo cost $1.98, all at 720p on Seedance 2.0.
In our three clips, the short brand words LUMO and EMBER stayed readable in every frame. Longer text was not tested.
Yes. Audio generation is on by default in Seedance 2.0, and all three of our files had an audio track. You can switch it off with generate_audio set to false.
Our Seedance 2.0 Fast run failed because of an audio copyright filter. Turn audio off when you add your own music and check your balance after a failed task.
We tested 720p. The documentation lists 480p to 4K for Seedance 2.0, with higher resolutions priced higher per second, so test at 720p and raise it for the final version.
