If you are judging an AI creative agency, or planning to make AI video ads yourself, the useful question is what happens before anyone opens a model. On our team, most of the thinking is done by then. We research, write a brief, and collect reference shots before the build starts.
This article walks through that order of work as we run it in November 2026. It also keeps the record of one earlier TikTok campaign, where CPA came in 50% below the prior creator-shot baseline. That campaign ran on an earlier setup, and its section says so.
The short version: We make an AI video ad in five steps: research, a brief, visual research, the build in our own software (voice, then images, then video), and revisions. An ad takes about a day. Expect a few revision rounds: up to 3 on our side, then up to 2 with you.
What is AI performance creative?
AI performance creative uses generative tools to produce image, video, and copy variants at volume under human creative direction. Teams then test the work against CPA, ROAS, and CTR. The approach combines fast production with a director who decides what each ad should say and why.
“AI ad creative” and “AI-generated ads” describe the output. AI performance creative describes the discipline behind it: the pipeline, the testing cadence, and the feedback loop between media performance and the next round of generation. That loop is what turns a batch of generated images into a creative system.
How do you make an AI video ad, step by step?
You make an AI video ad in five steps: research, a brief, visual research, the build, and revisions. A person owns the brief and the final decisions throughout. The build runs in our own software, which ties together the brief, the creation of assets, the first version of the ad, and the editing.
Five people work on our video ads. Here is how the steps divide.
| Step | What happens | Who or what does it |
|---|---|---|
| 1. Research | What is working, what is not, and what competitors are running | Our team |
| 2. Brief | The copy and the angle we believe will work | Our team |
| 3. Visual research | Reference pictures and the shots that should look good | Our team, with AI help |
| 4. Build | Voice (optional), then images, then video made from those shots | Our own software, running the models |
| 5. Revisions | Up to 3 rounds internally, then up to 2 with the client | Our team, then the client |
What research comes before the brief?
We start by looking at what is working and what is not. Depending on the client, we also look at what their competitors are doing. The aim is to find the angle before anyone writes a line or generates a frame, because a clear angle gives the model something specific to work toward.
Without that step, AI creative defaults to generic content. Our guide to competitor ad analysis covers how to turn what competitors run into your next creative test.
What goes into the brief?
The brief holds the copy and the angle we believe will work. Someone has to decide what the ad should make the viewer feel, and on our team that is a person. Every image and clip that follows is made to match the brief, so a vague brief produces a vague ad.
If the brief is “woman, 28, talks about loneliness,” the output will be flat. If the brief is “Korean woman lying in bed at 3am, hands over her mouth, embarrassed about how much she narrates her life inside her own head,” the output is a specific human moment. The model is the same in both cases. The brief sets the ceiling.
For the craft of writing one, see how to brief AI creative tools.
What is visual research, and how does AI help?
Visual research means finding reference pictures and deciding which shots should look good on screen. It is time-consuming, so AI helps us with it. The references we choose then guide the images and the video in the build, which is why this step comes before any generation.
What happens in the build?
The build runs in our own software, in a fixed order. We get a good pass at the voice first, then a good pass at the images, and then make the video from the shots we already created. The stills have to be right before the video starts, because the video is built from them.
Here is what we use at each stage as of November 2026:
- Voice (optional): ElevenLabs Eleven v4. Some ads do not need a voice.
- Images: mainly ChatGPT’s image tool, which has run on OpenAI’s Images 2.6 since November 4, 2026.
- Video: Seedance 2.5 first. When it cannot make a shot look real after several attempts, for example when the motion defies physics, we use Kling 4.0 for that shot.
Our AI video model comparison has the dated detail on both video models, how we get access, and what Seedance 2.5 costs per second.
How many revision rounds does an ad go through?
There are usually a few revision rounds: up to 3 internally, then up to 2 more with the client. The internal rounds happen before the client sees the ad, so the first version a client reviews has already been through our own edits.
What did the ads from the earlier campaign look like?
The ads below are from that earlier TikTok campaign. They were made on our earlier setup: the faces were generated in GPT Image 2 and animated in Seedance 2.0. Every character and every setting is generated. They are the ads behind the result in the next section.
Notice the skin under the train window light, the slightly uneven eyes, and the way the tie rests against the shirt. The character and train are both generated. Shot the usual way, this scene would have required a casting call and a half day of B-roll on the JR line.
A few more from the same run, with different settings and different talent:
What likely contributed to 50% lower TikTok CPA?
Three things most likely contributed: more variants per concept, more specific casting, and faster iteration. All three changed together on that one campaign, so we cannot separate them, and it was never a controlled test. The campaign ran on the earlier setup shown above, which means the current process has not earned that number.
Volume of variants. By our count on that campaign, the creator-shot pipeline produced 2 to 4 ads per concept because shooting was the bottleneck. The AI pipeline produced 12 to 20 because rendering is cheap.
Specificity of casting. With real creators we cast from who is available and who fits the budget. With avatars, the salaryman on the train is the one the brief described, in the lighting and wardrobe the brief described. That specificity is hard to buy on a creator marketplace.
Speed of iteration. When a hook tested well, we could ship a new variant of it the same day. When a hook stopped working, we replaced the creative the same day. The creator-shot pipeline measured iteration in weeks.
What does AI performance creative not replace?
It does not replace the creative director, performance judgment, or a true product story. The tools make production fast, and the result still depends on the people and the product behind them. Each of the three shows up in the work when it is missing.
The creative director. The work is stronger with a clear human point of view at the top. We once ran a sprint with AI driving the brief, and the variants felt generic. Putting the human director back fixed it.
Performance judgment. Knowing which hook to push, which avatar to retire, and when to rotate the whole concept set is still a paid media skill. Cheaper production leaves those decisions with people.
Product authenticity. The avatars work because the script underneath them is true to a real product and a real audience. Synthetic talent wrapped around an invented insight reads as hollow.
Can The Remarkable run this creative workflow for your team?
Yes. The Remarkable runs this workflow as a managed AI performance creative service. Human direction sets the concepts and hooks, production supplies the variants, and campaign results shape the next sprint. Our team reviews the work before launch and uses performance data to guide the next round. If video is the format you need, our AI video ads agency page shows the ads we have made.
If you are new to the term, start with what performance creative is and whether you need it. If you plan to hire for this work, how to brief an AI creative agency covers getting the first month right.
Bring your current ads and production bottleneck to a free strategy call. We will think through what to test first and which decisions your team should keep owning.