AWS · Computer vision · 2D AI video ad
AI Video Ad for Amazon Rekognition: Computer Vision Shown as Labels
AWS needed to show what Amazon Rekognition returns when it reads an image or a video, without asking viewers to know machine learning. This AI video ad turns each use case into a label a viewer can see: "SCHOOL BUS" on a street shot, a flag on a bad image, "SHIRT Style 5361" on a clothes rack.
Last updated
AWS · Amazon Rekognition
- Client
- Amazon Web Services (AWS)
- Product
- Amazon Rekognition, a managed computer vision service for images and videos
- Made for
- Developers and businesses that need to analyze images and video
- Format
- 2D line animation and live-action video
- Style
- Purple line icons on dark navy, label overlays on live-action
- Length
- 1:42
- Production timeline
- 3 to 4 weeks
- Budget
- $1,200 to $2,000
The problem: nobody can see what a computer vision service does
AWS describes Amazon Rekognition as a service that can "identify objects, people, text, scenes, and activities in images and videos", detect inappropriate content, and compare faces, with "no ML skills required". That is at least 6 capabilities, and none of them has a screen a buyer can picture.
A model is invisible. What a buyer can judge is the output: a label, a match, a flag. The video had to show that output for each use case in under 2 minutes, in one style, so the list reads as one product and not a feature dump.
How the video turns each use case into a visible label, scene by scene
Almost the only words on screen are the product name, "Custom Labels" and the labels the service would return. Select a time to play that scene above.
| Time | On screen | What it does |
|---|---|---|
| AWS logo, then a line-drawn person at a laptop: a clock ticks, images and video files pile up, a loading bar crawls and the screen scrambles. | Shows the manual work before naming the product. | |
| A magnifier icon and "Amazon Rekognition", then a folder, a stack of files and a strip of images pass through a scanning ring with a gear. | Names the product and shows it reading a whole library. | |
| Images and videos sort into rows by icon (people, car, plane), then one face is matched to the same face in other photos. | Shows classification and face search as results, not code. | |
| A live-action street shot enters the ring. Labels appear on its text: "SCHOOL BUS", "PARK ROW", "BROADWAY BUS LANES", "BUSES ONLY". | Proves text detection on a real video frame. | |
| A search bar over a list of images, then a red warning sign and a red box flag one item. | Shows content moderation as one flagged result. | |
| 3 people, each matched to an ID card with a check mark. | Shows identity verification. | |
| "Custom Labels", then a clothes rack tagged "SHIRT Style 5361", "PANTS Style 6498", "BOOTS Style 8265", a tagged logo in a social post, tags on a store aisle and on an assembly line part. | Shows labels only this business would need. | |
| AWS end card: "aws.amazon.com/rekognition". | Sends the viewer to the product page. |
Storyboard: the frames that carry the Amazon Rekognition AI video ad
Key frames from the finished video, in order. Select a time to play that moment above.
The pain first: a person, a clock and a loading bar. 
The product is named only after the problem. 
Classification as a result: the car row lights up. 
Face search: one face found in other photos. 
Text detection on a real street shot, with each source marked. 
Moderation as one flag in a list. 
Identity verification: each person matched to an ID. 
The custom feature gets its own title card. 
Labels a general model would not know: this store's style numbers. 
The same idea on an assembly line.
4 decisions behind the Amazon Rekognition AI video ad
Decision 1
Show the manual work before the product
From 0:04 to 0:11 a person sits at a laptop while a clock ticks, files pile up and a loading bar crawls. "Amazon Rekognition" appears only at 0:14.
Copy it: Spend the first 10 seconds on the job your buyer does by hand today. Name your product once that job is on screen.
Decision 2
Show the output, not the model
At 0:48 a real street shot carries the labels "SCHOOL BUS", "PARK ROW", "BROADWAY BUS LANES" and "BUSES ONLY", each tied to a dashed box on the sign or bus it read.
Copy it: For an AI product, draw what the model returns on top of a real input. Skip the neural network diagram.
Decision 3
One line style for every use case
Classification (0:26), face search (0:36), moderation (0:50) and ID checks (1:02) all use the same purple line icons on dark navy. Live-action appears only where the input is the real world: a street at 0:40, a store aisle at 1:28, an assembly line at 1:32.
Copy it: Pick one visual system before the script. Break it only when the real input is the point.
Decision 4
Give the custom feature its own card
At 1:12 "Custom Labels" fills the screen, then the rack shows labels no general model would know: "SHIRT Style 5361", "PANTS Style 6498", "BOOTS Style 8265".
Copy it: If your AI can be trained on the buyer's own data, show an example only that buyer would use, like their own part or style numbers.
From the Rekognition use-case list to a 1:42 line-art video
- Use cases to scenes. The capabilities AWS lists for Rekognition were mapped to one scene each: library analysis, classification, face search, text in video, moderation, identity checks and Custom Labels.
- Characters and environments. Line-drawn characters (the laptop worker, the matched faces, the 3 people at the ID check) and environments (desk, clothes rack, social post) were drawn in one purple-on-navy style.
- UI animation. The search bar, image list, warning flag and ID cards were animated as simple interface parts.
- Live-action and label overlays. Street, store aisle and assembly line shots were composited with dashed boxes and labels, so each detection sits on the thing it read.
- Motion graphics. A scanning ring with a gear and circuit lines links the scenes and returns for each new input.
- Edit, grade and master. The cut was edited, color graded, given sound design and mastered for output.
Scope of work:
Production took 3 to 4 weeks on a budget of $1,200 to $2,000. See the motion graphic video ad production service for how a project like this is scoped.
Questions about the Amazon Rekognition AI video ad
Why does the Amazon Rekognition video open on a person at a laptop?
It shows the job before the product. From 0:04 a clock ticks, files pile up and a loading bar crawls, and the screen scrambles at 0:10. The product name appears at 0:14, once the manual work is clear.
How does the video show text detection in video?
At 0:40 a real street shot enters the scanning ring. By 0:48 it carries the labels "SCHOOL BUS", "PARK ROW", "BROADWAY BUS LANES" and "BUSES ONLY", each linked to a dashed box on the sign or bus it came from.
Which Rekognition use cases does the video cover?
Library analysis (0:16), classification (0:26), face search (0:36), text in video (0:40), content moderation (0:50), identity verification (1:02) and Custom Labels (1:12), with retail, social, store shelf and assembly line examples after it.
How does the video explain Custom Labels?
With a title card at 1:12, then a clothes rack tagged "SHIRT Style 5361", "PANTS Style 6498" and "BOOTS Style 8265". Style numbers are labels only that store would need, which is the point of training on your own images.
Why does the video mix line animation with live-action shots?
The line icons keep 7 use cases in one style. Live-action appears only where the input is the real world: the street at 0:40, the store aisle at 1:28 and the assembly line at 1:32.
What did the Amazon Rekognition video take to make?
2D characters, environment illustration, UI animation, motion graphics, visual effects integration, sound design, editing, grading and mastering. Production took 3 to 4 weeks on a budget of $1,200 to $2,000.
What Advids clients say
"We used Advids to create an explainer video for an upcoming tradeshow. It was the first time we've worked on a project like this and the customer service experience was excellent throughout."Malcolm Treanor, COO at WINR Data
Need buyers to see what your AI model returns?
Talk to Advids