We use AI every day. Not to prove that we have our finger on the pulse. Not because a client asked us to. But because AI makes certain production processes faster, cheaper and in some cases better.
We also have clients who ask us to use AI for tasks it is not yet well suited to. And we have clients who think artificial intelligence cannot do things that it in fact masters extremely well. Both of these misconceptions cost money. Today’s post is my attempt at an honest account of where the boundaries of feasibility run today, in early 2026.
What AI does well
Language adaptations. This is the highest-value AI application in our pipeline. A master video produced in German can be adapted into English, French, Spanish, Italian, Polish and a dozen other languages using AI-assisted voiceover replacement, automated subtitle generation and on-screen text localisation. Each version is reviewed by a human. The total cost per additional language is a fraction of what a traditional dubbing and resync process costs. For clients with pan-European or global campaigns, this is a game-changer.
Background creation. For product shots, campaign imagery and visual composites, AI-generated backgrounds and environments have reached a quality level where they are production-usable for most applications. A product shot that would previously have required location scouting, a heap of set construction or expensive stock material can now be placed into a photorealistic environment more or less directly from a brief. The quality ceiling for hero campaign material is still below a dedicated shoot on location – but for secondary assets, social content and rapid iteration it is more than sufficient.
Retouching and image editing at scale. Removing backgrounds, adapting product colours, resizing and reformatting large asset libraries – these are tasks where AI has almost entirely replaced manual labour in our workflow. What a retoucher used to get done in two days now takes two hours – with human quality control, of course.
Rough outline and structure of a script. For explainer films and corporate content, AI-assisted first-draft scripting is genuinely useful – not as a replacement for a writer, but as a first structure that a writer then rewrites. It is faster than starting from a blank page, and it offers a variety of easily varied first options that a single writer might not have considered.
What AI cannot (yet) do well
Consistent characters and faces. We want to give it to you straight: even as I write this, the technologies out there are changing at a pace you can only trail behind whenever you try to sum up their quality. As of right now, character consistency across multiple images and scenes is possible, but image generation still comes with its share of snags and pitfalls. So if your campaign requires a specific person, a recurring character or a brand mascot to appear consistently throughout the video, that is still a challenge with generative AI and requires more effort than an unsuspecting AI user might assume at first glance.
Scientifically accurate visualisation. For life science and technical content, AI-generated imagery cannot be trusted to be accurate without extensive verification. A cell receptor shown in the wrong conformation looks right to a non-specialist but is wrong. AI generation optimises for visual plausibility, not scientific accuracy. For any content where accuracy is non-negotiable, AI assists the process but does not drive it. It remains essential that a human guides the image and video generation attentively and that the result is checked closely.
Brand-specific visual language without custom training. Generic AI models produce generic results. If your campaign needs to look like your brand – colours, typography, compositional style – a generic model will only approximate it badly. Brand-consistent AI output requires either custom-trained models or extensive human direction, iteration and post-processing. We do all three, but none of them is “fast and cheap” in the way the AI hype suggests.
Replacing creative direction. AI is a production tool, not a creative director. The brief, the strategy, the narrative structure, the decision about what to show and what to leave out – all of this requires human judgment that AI cannot currently replace. The studios that treat AI as a replacement for creative thinking are producing content that looks like it was made by AI – which for viewers is currently a clear signal of low quality, even if they cannot articulate why.
The bottom line
AI makes the adaptation and distribution layer of production dramatically more efficient. It does not yet make the creative and production layer significantly faster for high-quality output.
The studios claiming otherwise are usually showing you only the best 5% of AI-generated frames – not the production output that took three hours of prompting, waiting, reworking, combining, prompting, waiting again and so on.
In our experience, clients today assume that far more is possible than just a year ago – and they are right about that. At the same time, most of our clients have already had a go at prompting on their own, perhaps hoping to get a job done quickly and without outside support. In the process they have had to learn the painful lesson that it is not enough to simply press a button and *snap* – all audiovisual wishes come true. Batting an already quite decent first attempt back and forth until the end result actually matches your own vision and is genuinely worth showing takes hard work and a lot of know-how. We bring both, complemented and brought to full bloom by traditional tools and techniques that balance out AI’s weaknesses and make the process controllable. We use AI where it genuinely helps. Where it does not help, we do without it. That distinction is a decisive part of our work.
