March Insights

By Eve, Artstash Creative · Published 25 March 2026 · Last updated August 2026

Quick answer: Three things moved in March. AppLovin’s Axon began rolling out a fully automated multi-agent creative pipeline, generating thousands of playable interactives for hundreds of clients in a matter of weeks. Runway showed its Gen-4.5 video model can pass for real footage in a blind test of 1,000 viewers. And cinemagraphs are quietly posting higher conversion rates again. The common thread: automated generation is now cheap enough that creative differentiation moves back to strategy and format choice.

How Is Axon Automating Creative at Scale?

In a recent blog post, AI marketing platform Axon by AppLovin described their recent rollout of a fully automated, multi-agent production pipeline for ads on their platform. Axon ads consist of three main parts; a video, interactable HTML element, and a dynamic catalog view, with the interactable HTML being the biggest pain point for clients. Selected advertising partners can now opt into receiving custom interactive HTML layers within minutes.

Example of generated interactable for David protein bars. Source: Axon

The platform is made up of different AI-powered agents whose roles are:

• Brand context, product imagery, and messaging guardrails are gathered automatically
• Multiple structured concepts are drafted
• Creative assets are generated and iteratively refined using image and video models
• Quality checks are performed before delivery

After just a couple of weeks, thousands of interactives were generated for hundreds of clients, all with very positive feedback:

However, this was only a test of the feature. The anticipated rollout is expected in early Q2 where Axon plans to increase the breadth and depth of this pipeline, with the goal of “[automating] the full creative stack”. They even aim to produce AI generated playables for their clients in the gaming industry.

Can Viewers Tell Runway AI Video From Real Footage?

Runway recently showcased their newest Gen-4.5 base model for video generation in a unique way by proving that viewers couldn’t tell it apart from real life.

During a survey 1,000 participants were shown a series of 5-second clips, 50% of them captured in real life, and 50% of them generated from a single image with no regeneration or editing. Only 99 of the participants could distinguish between real and generated footage. They have since published a quiz to test your own ability to tell the two apart – how did you do?

Are cinemagraphs making a comeback?

We are seeing a noticeable increase in CVR of cinemagraphs ads specifically for casual games; simple designs with no flash and very little movement:

Example of a cinemagraph. Credit: World Winner

Our hypothesis: if a viewer’s attention is interrupted while watching an ad, the video continues on and we lose the chance to convert the viewer. However, in the case of a looping cinemagraph, the content remains on screen until the viewer scrolls or engages. While it’s too early to determine whether these are yielding long term definitive results, we should be challenging as to whether this could increase the likelihood of viewer engagement and conversion?

What we can say is that cinemagraphs are quick to create and iterate at scale, as well as having cheaper production costs than a video. Therefore, it would be worth exploring as an additional execution type to test in your next campaign.

If you’ve experimented with using cinemagraphs in your creative for casual games we’d love to hear your performance experience in the comments.

Frequently asked questions

What is Axon by AppLovin doing with automated creative?

Axon is rolling out a multi-agent pipeline where separate AI agents handle distinct roles in producing playable and interactive ad units. In an initial test it generated thousands of interactives for hundreds of clients within a few weeks, with a broader rollout expected in early Q2.

Can people tell AI-generated video from real footage?

In Runway’s test, 1,000 participants were shown a series of five-second clips, half filmed and half generated by its Gen-4.5 model. The results indicated viewers could not reliably distinguish the two, which is the point the company was making about model maturity.

Are cinemagraphs worth testing again in mobile UA?

We are seeing a noticeable increase in conversion rate from cinemagraph ads. They sit between static and full video, so they are cheap to produce relative to video while still arresting attention in feed, which makes them a low-risk format to add to a test matrix.