Sora went dark in April 2026. Sixty days on, every UK marketing team that had built a slide into their 2026 plan around it is quietly rebuilding the stack underneath. If you're a brand video buyer right now, the practical question is not "which AI tool is best", it's "which two or three should we standardise on, and what do we keep in human hands".
This is the studio view from the inside, written for the person who has to make those calls before another quarter slips.
The post-Sora landscape, in plain terms
When OpenAI pulled Sora it wasn't a graceful sunset. The tool was running at roughly fifteen million dollars a day in compute against lifetime revenue of about two million. The exit reshuffled the enterprise stack in a single weekend and left three credible contenders for the workload UK brands were sending to it: Google's Veo 3.1, Synthesia and HeyGen for avatar-led work, and Runway plus Kling for the more directed generative jobs.
Veo 3.1 is the one most UK marketing leads will land on first, and for good reason. It outputs genuine 4K up to 60fps. Native audio means it generates dialogue and ambient sound in the same pass, not bolted on. And the deep Vertex AI and Google Cloud integration is decisive for any team already running on Google Workspace, which is most of them. Synthesia and HeyGen are the workhorses for talking-head avatar pieces in twelve to thirty languages, which is where a lot of internal comms and training video lives.
The market data behind the reshuffle is consistent. Roughly seventy-eight percent of marketing teams now use AI-generated video in at least one campaign per quarter, up from thirty percent in early 2024. Average production timelines on a sixty-second piece have collapsed from thirteen days to twenty-seven minutes when the workflow is set up well. Production cost on the same brief is down ninety-one percent against the traditional approach, with eighty-one percent of marketers reporting at least three hours saved per project.
Those numbers are real. They are also misleading if you treat them as the whole picture.
Where the AI stack genuinely shortens the pipeline
There are three places it earns its money for a brand video buyer.
The first is concept and ideation. Veo and the rest can turn a half-formed creative brief into ten or fifteen rough visual treatments inside a working hour, which gives a marketing director something concrete to react to before any production money is committed. Used this way, AI is a thinking tool, not a production tool, and that is its highest-leverage use.
The second is the long tail of localisation. Once a hero film is in the can, generating fifteen language variants used to mean a fresh studio day, a translation pass and a re-edit per market. Synthesia and HeyGen now do the bulk of that work overnight, with a human pass for the markets that genuinely matter to the campaign. The cost saving on a multi-market roll-out is the single biggest line-item shift we have seen in the last twelve months.
The third is the unglamorous production work most clients never see. Rough-cut assembly, captioning, asset versioning into fifteen format variants for paid, social and web. AI is already doing that on most large UK accounts. It happens behind the scenes, it does not threaten brand voice, and it frees the editor to work on the things that actually move the needle.
Where it introduces rework
The same three categories of work that AI is bad at have not changed in eighteen months. They are worth naming because they are where most of the AI stack horror stories come from.
Hero brand pieces with real people in them. The eighty-percent voiceover market shift to AI does not extend to the founder telling the story, the customer giving the testimonial, or the engineer explaining what the product actually does. Audiences disengage from those moments fast when they are obviously generated. Seventy-eight percent of consumers trust videos with real people more than AI-generated content. That number is even higher among the more sceptical UK audiences and (a point worth weighing for any team running both sides of the Atlantic) South African audiences tune out known-AI content at the highest rate in the world.
Complex motion with specific real-world physics. Tools have closed the uncanny-valley gap faster than anyone predicted, but they still hallucinate on the details that matter to a brand: how a bottle pours, how fabric moves, how a logo sits on a moving surface. If your campaign needs the product to behave exactly as it does in real life, you are still better served by a shoot day or a small VFX pass.
Anything that requires editorial judgement under pressure. Cutting a thirty-second spot from a three-hour edit is a craft decision, not a length decision. The AI cuts are coherent. They are not interesting. The difference shows on a brand's third or fourth piece of the year, when the audience starts to feel the pattern.
What a brand asset library needs to look like
This is the part most UK teams under-invest in and then regret six months later. AI tools are only as on-brand as the inputs you give them. A library that was good enough for human designers and editors is rarely structured enough for a generative tool to draw from cleanly.
What works in 2026: a single source of truth for brand assets, with every file tagged in plain language (not just by name), versioned, and held in a system the AI stack can read from. That usually means a DAM with an open API, brand guidelines written as plain text and not just PDF, and a small library of approved reference clips that show the brand in motion. Many of the bottlenecks teams blame on the AI tools are actually bottlenecks in the library feeding them.
If you do nothing else this quarter, audit your brand library against the way your AI stack reads. That single move pays for itself inside a campaign cycle.
The calls worth making before another quarter slips
Pick one generative tool and one avatar tool. Trying to keep up with four parallel stacks is how teams burn budget and end up with inconsistent output. Veo 3.1 plus Synthesia or HeyGen is the most common shape we see working in the UK market. Pick yours, train the team, and resist the urge to swap quarterly.
Write your AI policy into the brief, not the contract. The brief is where decisions get made about the work; the contract is where they get formalised. If the brief does not say where AI is acceptable in this project and where it is not, you will discover it in the edit, which is the most expensive place to have that conversation. We unpack this in the brief that saves the project.
Keep the hero film, the founder, and the customer story in human hands. Use the AI stack to amplify what they produce, not replace it. That balance is what holds brand trust together as the rest of the pipeline gets cheaper and faster.
A studio note to close
This is the conversation we are having with most of our UK brand clients right now: the stack matters less than the discipline around it. We run Veo for concept and localisation, our editors for the hero work, and a strict brief that names where each lives. We would rather over-spend forty minutes on the brief than save forty hours in post and lose the brand voice. The studios still worth hiring in 2026 are the ones who can hold both halves of that line at once.
Create With Purpose.
The Creative Clan
Cape Town • London
www.creativeclan.net
