A dynamic video ad on Meta is two jobs that people tend to blur into one. The platform assembles and serves ads from a catalog. Someone else has to produce the video that goes into them. We do the second job, and this page is about how to do it from a product feed without turning your motion team into a copy shop.
The reader we have in mind is the producer or media buyer at a small agency or in-house team. You have a catalog of products in a feed. You have a media plan that wants more than one video angle per product, in more than one shape. And you have one motion designer, who is already busy. The question is how many finished files that designer can responsibly hand over, and who decides which ones exist.
What the platform does and what a render does
Meta’s catalog-driven ad formats pull product details from a feed and put them into an ad layout. We are not going to describe that machinery in detail, because it changes, because it belongs to Meta, and because we do not make it. Read Meta’s current documentation for what each format accepts, what video specifications it expects, and how it picks which item a viewer sees. Treat anything we say about the platform as general practice and check it against their pages before you build around it.
What we can say is where our work starts. Templater is an After Effects plugin that merges a rigged template with a data source and renders or replicates one composition per row. The data source can be a Google Sheet, a tab-separated file exported from Excel, Numbers or Sheets, or a JSON file or feed. So the feed that describes your products can also be the work order for a set of video files. Each row becomes a finished video, with the product name, photo, price line and badge pulled from that row. If you want the longer version of that mechanic, our guide to making product or listing videos from a data feed walks through it.
Put the two jobs side by side and the division is clean. The platform decides what gets shown to whom, and when. The render decides what the video looks like and says. Our product pages list Vimeo, JW Player, Frame.io and AWS S3 as the places Projector sends finished videos. They do not list advertising platforms. That is worth knowing early, because it means the step from a finished file to a live ad is yours, whether a person uploads it or a script does.
Here is the practical consequence. A video made from a feed is a set of ordinary video files. It is not a live connection between your catalog and the platform. If a price changes after the render, the file still shows the old price until you render it again. If your priority is a creative that updates itself every time the catalog changes, the platform’s own catalog format is built for that, and a rendered video is a different tool. If your priority is a polished motion-designed video for each product, with timing and type you control, rendering from the feed is how you get it. Some teams want both, a catalog format for the long tail and rendered video for the products that carry the budget. That split is a reasonable one, and nothing in our tools stands in its way.
A picture of the job
Take a regional furniture retailer, offered as an illustration and not a customer. The feed has a few hundred products, each with a name, a photo, a short feature line, a price line and a category. The marketer wants video for the products that sell the most, and the media buyer wants to learn which opening line works. The designer has one master composition for a tall placement and one for a wide one.
Nobody should render a few hundred products times every angle times every shape on day one. Most of that output would never see a viewer, and every file is something to check, name, upload and eventually retire. The sensible first move is to pick a slice. The top products by sales, in one shape, with two or three angles each. That is a few dozen files, enough to learn from, and small enough that a person can look at every one before it ships.
Plan the variants as columns, not as copies
This section is general practice, because it is about how to think and not about a setting. A common mistake is treating each variant as a separate project. Someone duplicates the template, changes the headline, saves it under a new name, and does it again. By the tenth copy the team cannot say which one holds the latest logo.
The alternative is to decide what varies and give each thing a column. The product name, photo and price line come from the feed. The angle, meaning the opening line or the benefit you lead with, is a column you add. The shape is a column too, if you use more than one master composition, because Templater can switch between template projects from the data source on a per-row basis. One sheet can then drive a tall version and a wide version of the same product, each built on its own layout.
That last point deserves a sentence of caution. A wide composition and a tall one are separate layouts. Things that sit side by side in one will stack in the other. Layout rules help elements follow their containers and sibling layers, so a price stays below a name when the name runs long, but they will not redesign a wide layout into a tall one. Build each shape on purpose, as its own master, and test both against your longest product name before you call either finished.
The angle column is where the media buyer earns their keep. A short list of angles written by a person who understands the shopper beats a long list that a script produced. Three angles that differ in what they say will teach you more than ten that differ in a word. This is the part of the work we cannot do for you, and we will not pretend otherwise. We make the file. We cannot tell you which opening line will move your audience, and we have no numbers to offer on what a good variant count is. Your own results will have to answer that.
One more column pays for itself, and it is the file name. Templater can set a custom output file name for each job from the data source. If every file is named from the product identifier, the angle and the shape, then the name tells you what the video is without opening it. That helps the person who uploads it, and it helps the person who reads the results later and needs to know which angle a number belongs to. A folder of files called final and final-2 hides exactly the information a test needs. Decide the naming pattern before the first render and write it down.
Exceptions belong in cells
Catalogs are never uniform. One product has a sale badge and the rest do not. One category needs a line the others do not. If each exception becomes its own copy of the project, you are back to the pile of duplicates. Templater supports flags in the data source, which let a cell act as a switch inside the template, so an exception is a value in a row and not a fork. Our post on using flags in data sources explains how they work, and it is worth reading before you design the columns, since it changes how many you need.
Text is the other place that bites. A single text layer takes one font, one size, one weight and one style, so a headline with one bold word in the middle needs two layers positioned side by side, not one marked-up string. Plan the type for the longest product name in the feed, not the typical one, because the longest one is the row that will break the layout on the day you least want to see it.
Check, name, hand off, and retire
Rendering is the quick part. What surrounds it is where ad teams lose time, so it helps to treat the pipeline as four plain stages.
First, check before you commit the machine. Templater can preview compositions with your data before it renders or replicates, and Pro and Bot can export stills or short clips of chosen frames from the merged compositions. That gives you a contact sheet of the products with the long names, the missing photos and the odd prices while a fix is still cheap. We cover the choices in our post on preview modes and spot checks. For a feed of unknown quality, look at the worst rows first. A clean feed passes quickly, and a dirty one tells you something about the work ahead.
Second, render. The Pro edition renders when someone clicks. The Bot edition checks periodically for incoming data and starts the work on its own, as a background process, and it can be driven from the command line. A team making a defined slice once a week can live with Pro. A team whose feed changes daily and whose catalog is large may want Bot. Bot reads a render-status column in the data source to decide whether to render a row, and it stops with an error if that column is missing, so add the column before you start. If you want to feed jobs in from other systems, QUE is our JSON data manager with an API that receives data from web and mobile apps, and a Zapier integration. Our QUE page has the details. And if you are weighing the products against each other, Templater is the one that renders.
Templater editions
Which edition fits your workflow
| Edition | How it triggers | Key capability |
|---|---|---|
| Pro | Someone clicks to render | Exports stills or short clips for spot checks |
| Bot | Checks periodically, runs as background process | Reads render-status column; can be driven from command line |
| QUE | Receives data from web and mobile apps via API | JSON data manager with a Zapier integration |
Estimate the time before you promise a delivery date. Rendering time depends on the length of the videos, the output resolution, your hardware, the ratio of dynamic to static frames, and how heavy the effects are. We cannot give you a figure, because it is yours to measure. Render one representative video on a machine like the production one, multiply by the count, and compare that to the time you have. A second machine running Bot can divide the work, and the data source has to be set up for multiple Bots.
Third, the handoff. A finished file is not yet an ad. Someone has to upload it to the ad account, attach it to the right ad set, and keep the name. Templater supports event scripts, which are shell scripts and ExtendScripts that hook into its processes, and the FAQ describes a script that uploads videos once they render and writes the resulting URL back to the data source. We do not have a documented connector to Meta, so if you want the upload automated you are talking about a script, and our custom development team can quote one. For many teams a person dragging a folder into the ad account is fine at this scale, so long as the file names carry the meaning.
Fourth, retirement, which is the step everyone forgets. A product goes out of stock, and its video keeps running. A price changes, and the old one stays in the video. Because a rendered file is a snapshot, the retirement habit has to live in your routine and not in the tool. Keep the product identifier in the file name, so that when a row changes status you can find every file and every ad that came from it. We wrote about this at more length in keeping video ads current when inventory turns over, and the same rules apply here.
Where this is the wrong approach
Rendering variants from a feed is a poor fit in a few cases, and it is better to say so now.
If your whole catalog changes by the hour and you cannot accept any gap between a change and a new video, a rendered file will always lag. The render takes time, and so does the handoff. The platform’s own catalog format exists for that situation, and you should use it there.
Two tools, two jobs
Catalog format vs. rendered video: different tools
- Updates when feed changes?
- Platform catalog formatYes, live view of the feed
- Rendered video from a feedNo, file shows old data until re-rendered
- Creative control over timing and type?
- Platform catalog formatPlatform controls the layout
- Rendered video from a feedYou control timing and type
- Best for which products?
- Platform catalog formatLong tail, constantly changing catalog
- Rendered video from a feedProducts that carry your budget
An account can reasonably use both.
If you have a small number of hero products and one strong idea per product, a motion designer making each video by hand may well produce better work than any template, and you will not need a data pipeline at all. A template standardizes. That is its strength and its limit. When each product needs a different story, a different edit and different pacing, the data-driven approach will feel like a straitjacket.
If the goal is to discover what works with a wide spread of cheap tests, make sure you are testing something that matters. A hundred variants that differ by a color will teach you little and cost you a lot of checking. Fewer variants that differ in what they say will tell you more. We would rather you render thirty files you understand than three hundred you cannot account for. For help thinking about versions at a larger scale, our post on ad versioning in After Effects goes into it, and making hundreds of variations from one template covers the high-volume end.
There is also the question of what we will not do for you. Templater does not choose your products, write your angles, manage your ad budgets or audiences, or report results. It makes the video files from the rows you give it. The feed has to be clean, because a wrong price in a row is a wrong price in a video, and it arrives at the speed of automation.
Who does what
A workable split looks like this. The designer builds and rigs the master compositions once, one per shape. The marketer or media buyer owns the sheet, meaning the angles, the file naming and the list of products in the slice. A producer runs the check, the render and the handoff. Over time the designer touches the master only when the brand changes, and the sheet becomes the place where campaigns are planned. That is a calmer arrangement than a designer pasting headlines into a timeline, and it keeps the creative decisions with the people who made them.
To put the bottom line plainly. Meta’s catalog formats and a rendered video from a feed do different jobs, and an account can reasonably use both. Render from the feed when you want designed, motion-quality video for the products that matter, in a deliberate set of angles and shapes. Start with a small slice, name every file from the product, the angle and the shape, look at the worst rows before you render the batch, and give retirement an owner. Check the platform’s own documentation for what it accepts, because that part is theirs and not ours.
We cover the rest of the pipeline in the pages linked above, and our learning hub collects them in one place.
Before you plan your first slice, here are the questions teams ask us most about this kind of work.
Frequently Asked Questions
Does Templater connect directly to Meta’s ad platform?
No, and nothing on our product pages says it does. Templater reads a data source and renders or replicates videos. Projector sends finished videos to Vimeo, JW Player, Frame.io and AWS S3. Getting a file into an ad account is a step you handle by hand or with a script. Our custom development team can write scripts that hook into Templater’s events, and engagements are quoted individually. For what Meta accepts, read Meta’s own documentation.
Can the product feed be the data source for the videos?
If the feed can be a Google Sheet, a tab-separated file, or a JSON file or feed, yes. Templater reads all three. It cannot read an Excel file directly, so export that to a tab-delimited file first. It also maps to one worksheet and one data source at a time, so put the whole batch on one tab. If the feed arrives in another shape, a script or a sheet that something writes to is the usual bridge.
How do I make a tall version and a wide version from the same sheet?
Build a separate master composition for each shape and add a column that tells Templater which project to use. It can switch between template projects on a per-row basis from the same data source. Treat each shape as its own design, because layout rules will keep elements in relation to each other but will not turn a wide layout into a tall one.
How many variants should I render per product?
We do not have a documented number, and we will not make one up. General practice is to begin with a small slice, the products that carry the most budget, and a few angles that differ in what they say. Look at your own results and add from there. Every file is something to check, name, upload and eventually retire, so the count has a real cost on your side.
What happens to a video when the price in the feed changes?
Nothing, until you render it again. A rendered video is a file, not a live view of the feed. Put the product identifier in the file name, using Templater’s custom output naming from the data source, so you can find every file and ad that used the old price. Bot reads a render-status column to decide what to render, so test how it treats a changed row in a rehearsal before you depend on it.
Is there a way to check the videos before rendering the whole batch?
Yes. Templater can preview compositions with your data before it renders or replicates. Pro and Bot can also export stills or short clips of chosen frames from the merged compositions, which is a fast way to see the product with the longest name or a missing photo. Look at the worst rows first, and fix the template or the data before you commit the machine.
When should I use the platform’s catalog format instead of rendered video?
When the catalog changes constantly and you cannot accept the gap that rendering and handoff create, or when you want a long tail of products covered without a designed video for each. Rendered video from a feed suits the products that carry your budget and deserve controlled timing and type. Some accounts use both. Check Meta’s documentation for the current specifications of each.
