Creative testing burns through visuals faster than a design team can produce them. A concept that works still has to be tried against other backgrounds, other faces, other products and every placement, and each of those is a separate file. We make AI ad creatives for exactly this step: you bring a concept or a finished creative that is already approved, and we turn it into ad creative variations for testing, still images and short clips, while keeping what defines the ad intact.
The work is done by our team, not by a self-serve tool. We set up the generation, show you a test batch, run the full set on our GPUs after your approval, review every file by hand and deliver the set as a ZIP archive or a link to a shared folder. You decide what goes into the test and what the ads say. We make the pictures and the clips.
One concept, many ad creative variations
Useful variations change one thing on purpose. Before the run we agree on which parts of the creative are fixed, such as the composition, the product, the brand colors and the space for copy, and which parts are allowed to move. Variables worth testing:
| What changes | Example |
|---|---|
| Background and setting | the same product on a kitchen counter, in a car, on a beach |
| Scene | morning routine, office desk, gym bag, gift unboxing |
| Character | different ages, looks and outfits, or your own mascot in new poses |
| Product | other colorways, flavors or SKUs from the same line |
| Color and light | warm or cold palette, daylight or evening light |
| Season | summer, autumn, winter, holiday decor |
| Angle and framing | close-up, mid shot, top-down, the product turned the other way |
You can test one variable at a time or combine them, for example three backgrounds × two characters × two formats, which already makes twelve variants. Combinations multiply fast, so we fix the grid before the bulk run, not halfway through it.
Static images and short AI video ads
Static variants are made in one of two ways, depending on the brief.
- Editing your approved creative. For a new background, season, time of day or color, we work from the approved image itself with reference-based models such as FLUX.1 Kontext [pro] and Nano Banana. The original layout stays the starting point, so the variants still read as the same ad.
- Running a template. When you need many new scenes or characters in one style and you have a ComfyUI workflow, we run it with the checkpoints and LoRAs it specifies: SDXL, Flux or others. This is the same process as our bulk image generation.
Short video variants are made with image-to-video. A still, either your approved creative or one of the static variants, becomes the first frame, and the motion is described in a prompt: a slow push-in, a model turning toward the camera, steam rising from a cup. Our production graph runs WAN 2.2 image-to-video, interpolates to 32 fps, upscales the result and delivers an H.264 MP4. At the base price a clip is short, about 5 seconds; longer clips or higher resolution are quoted separately. If you do not need a template and just want the result, we can make the clips with commercial video models via API, such as Veo, Seedance or PixVerse.
Two limits are worth knowing before you plan AI video ads with us. Image-to-video needs a start image: no still, no clip. And small text and logos are the parts most likely to drift once the frame starts moving, so headlines and disclaimers are better added in editing after generation. The video side is described in detail on the bulk video generation page and in the guide on WAN image-to-video at scale.
Formats: 1:1, 4:5, 9:16, 16:9
We deliver in the four aspect ratios most ad placements use: 1:1, 4:5, 9:16 (1080×1920) and 16:9. For video, vertical 9:16 is our working format; other ratios are available on request.
Name your placements in the brief. A composition built for a vertical story does not always survive a crop to a square: the product ends up at the edge of the frame or the face gets cut off. When the formats are known up front, the test batch includes every ratio you need, and you see the problems before the bulk run, not after it.
If your designer adds the copy, tell us where the headline and the button will go. We can deliver clean backgrounds with that area left calm, ready for layout.
Keeping the brand recognizable
A variant is useful only if the ad still looks like your brand. Three things keep it on track.
- A LoRA for what has to stay the same. If a character, mascot or product appears in every variant, we can train a LoRA on it from 5–25 photos, from $99 one-time. The model then keeps that particular face or package recognizable instead of inventing a lookalike. Which photos work best is covered in the guide on character LoRA datasets.
- Catalog shots as the base. If you sell many SKUs, the shots from AI product photos also work as starting frames for ad variations, so the product in the ad matches the product in the catalog.
- A test batch and manual selection. Before the full run you get a small batch and either approve the direction or ask for changes. After the run we review every file by hand and drop the broken ones: extra fingers, warped packaging, a melted logo, a face that drifted away from the character.
How many variants to test
There is no universal number. It depends on your budget per test, on how many ad sets you can run in parallel and on how quickly each one collects enough data to judge. What we suggest is a structure rather than a figure.
- Decide what the test should answer. “Which background works” and “which character works” are different tests. Changing everything at once gives you a winner but not the reason it won.
- Start with a modest round. A few variants per variable across the formats you actually run are enough to see whether a direction holds. The next round goes deeper on what did well.
- Keep a variant map. Note what changed in each file, so the results in your ad account can be traced back to a specific background, character or format.
We do not promise a higher CTR, a lower CPA or a better ROAS. What we change is how many options you can afford to put into a test. Which option wins is decided by your audience, not by us.
What we don’t do
If you are looking for a full-service AI ad creative agency with strategy, copywriting and media buying, we are one part of that chain: the production of visual variants. Outside this service:
- Concepts, copy and offers. We work from an idea you have already approved. Headlines, offers and calls to action are written on your side.
- Text on the creative. We do not set headlines by default. If you send a finished layout with the copy in it, we discuss case by case how to keep it; otherwise you get clean visuals with room for text.
- Talking avatars and lip-synced UGC. No AI presenters, no lip-sync, no voice-over or voice cloning.
- Media buying. We do not launch campaigns, manage ad accounts or analyze campaign results.
- Guaranteed ad approval. Ad review is up to each platform, and we cannot promise that a creative will pass it.
- Adult content.
Pricing
- Images: from $0.10 per variant.
- Short video variants, about 5 seconds: from $1.50 per clip. Longer clips and higher resolution are quoted separately.
- LoRA of a character, mascot or product, if needed: from $99 one-time.
- Setup and the test batch are included. The minimum order is $50.
These are starting prices. Because static variants are inexpensive, a small image-only round can come in under the minimum, so it often makes sense to combine statics, clips and several formats in one order. Price, payment and timing are agreed after the brief and the test batch.
Send us the creative
Send the approved creative or a short description of the concept, what should change and what must stay, the formats you need and a rough volume. You do not need your own GPU or any ComfyUI skills. Fill in the form and we will come back with questions or a plan for the test batch.
