Tech • AI • Robotics • Game
We tested the new Kling 4.0 on Higgsfield in a real ad (the result?)
5/10TL;DR
Xfield and Kling 4.0 were used to build a longer AI-generated ad for Tapit, highlighting both the promise of low-cost video production and the current limits of realism, continuity and brand control.
KEY POINTS
From 10 seconds to 30 seconds
The project moved from a short 10-second concept to a 30-second commercial using Kling 4.0, whose main upgrade over the previous version was longer clip generation. The aim was to turn a simple visual of a creator receiving sales notifications into a more complete advertising narrative with a beginning, middle and ending.
A revised storyline for Tapit
The ad concept centered on a digital creator promoting a new Tapit store product and transforming attention into income. The first idea showed a creator on a terrace in Bali, checking a phone as payment alerts appeared, then ending in a scene of freedom and success at sunset.
Push for a darker, more realistic opening
Feedback on the first 30-second result focused on realism. The creative direction shifted toward a colder, more subdued opening with multiple short shots, including rain, low energy and a sense that “everything is going wrong,” before the turnaround. One proposed image showed the creator with 20,000 followers but €0 in revenue, underlining the gap between audience and monetization.
Targeting a European digital nomad
The character was redefined as a European digital nomad rather than a local figure in Bali, reflecting the campaign’s intended customer profile. The stated objective was to align the ad more closely with the expected audience for online store tools and avoid a generic or mismatched visual identity.
Single-shot output judged too weak
The first generated ad was criticized for relying too heavily on a single tracking shot and an overly artificial smile. The team wanted a sequence of back-to-back shots lasting roughly three seconds each, creating a stronger emotional arc from frustration to financial freedom rather than a single smooth camera move.
Brand assets and interfaces needed improvement
Another major weakness was the lack of believable on-screen interfaces. The payment alerts and screen designs were seen as unconvincing, prompting a request for more realistic, Apple-style notifications and direct use of real product assets such as logos, store visuals and design files.
Frames and references as control tools
To improve consistency, the workflow relied on two features inside Xfield: frames and references. Frames allowed the ad makers to set key moments such as the opening image, mid-point reveal and final shot, while references enabled direct uploads of logos, product designs and screenshots for the model to incorporate into the video.
Results showed progress but clear flaws
A second attempt produced a structure that was considered more coherent at the beginning and end, but the middle transitions and product integration still fell short. The experiment showed that AI could roughly follow an intended narrative, but stitching scenes together and maintaining visual quality remained difficult.
Low-cost production remains a central pitch
Despite the flaws, the broader commercial argument was that AI tools such as Kling 4.0 and Xfield can sharply cut production costs. The comparison put an AI-generated creative at around €29 to €39, versus a traditional agency production that could run from €2,000 to €10,000 or more for a comparable ad concept.
CONCLUSION
The experiment underscored both the rapid progress and the unfinished state of AI video advertising. Kling 4.0 and Xfield can already generate structured, brand-guided commercials at a fraction of traditional costs, but realism, transitions and emotional credibility still require substantial iteration.
Ask a question