Testing and R&D
We give the tool to somebody who has not learned to work around it.
The test room already tests cameras, stabilisation systems, drones and virtual production kit with manufacturers and research institutions; AI tools go through the same method. Most published evaluation is done by experts who have learned every workaround before they start timing anything. We test with new entrants and working crew on live units, where a failure costs an hour of a shooting day.
Last checked September 2026.
How a test runs
How a test runs.
01
Scope and terms
Agreed in writing: what is being tested, on what kind of work, who owns the material produced, and that the result is published whatever it says.
02
Set the task
A task from a production or simulation, defined before anybody touches the tool, with the manual baseline recorded.
03
Run it on the day
The tool goes to crew in the grade that would use it, on a working unit, under the pressure of a schedule. Everything that goes wrong is recorded, including the things caused by us.
04
Publish
What we asked, what happened, where it failed, what it changed about the day, and what we would now use it for. The developer sees it before publication and may correct facts, not conclusions.
What gets published
Enough for somebody else to disagree with us.
A test report says what the task was, what kit and version were used, who did it and at what grade, what the tool did, and what the failure modes were. Results unfavourable to the developer are published, and so are results unfavourable to us.
- List item text
- List item text
For manufacturers and developers
What working with us looks like.
Your tool in front of the users you never see: people entering the industry, and working crew who have not been trained by your team. Failure modes from a shooting day, and a written report you can act on. We ask for a defined scope, the right to publish the result, and a fair commercial arrangement for the days involved, confirmed in writing before anything is booked.
Why this, and why now
The UK provision we can verify is literacy and law. None of it publishes measured results from a working unit.
ScreenSkills has accredited its first AI training course, delivered by the consultancy AIMICI with Bournemouth University and UCL, and described as building durable skills, practical frameworks and best practice drawn from commissioner, broadcaster and union guidance. ScreenSkills chief executive Laura Mansfield framed the accreditation as addressing employer uncertainty about how AI will affect specific roles.
The report we read carries no date, so we do not date it. The rest of the funded UK provision we could verify runs the same way: a ScreenSkills course on AI for production roles in scripted television and film, delivered by a media lawyer and covering platform codes, regulatory risk, IP and consent; a ScreenSkills AI fundamentals course; a development toolkit; and a Film Skills Fund course with StoryFutures at Royal Holloway and Final Pixel. All of it is literacy and law, and all of it is worth doing. None of it publishes what happened when the tools were run on a working unit, which is the gap this Lab exists to fill.
Scope
What we can put on a bench.
| Category | What a test looks like |
|---|---|
| Generative image and video | A named task with a defined output, run to a brief with acceptance criteria written before the tool is opened, and scored on whether the result could be handed to the next department. |
| Post, sound and restoration | Repair, separation, upscaling and conform against material we shot ourselves, so the ground truth is known. |
| Prep, breakdown and scheduling | A script into a schedule, measured against the same job done conventionally, with the mis-tags and omissions counted. |
| VFX, 3D and asset generation | Character work, asset creation and rigging on plates with known difficulty: contact, occlusion, crowd, motion blur. |
| On-set assists and logging | Whether it survives a working day: time pressure, poor connectivity, a changed brief, and somebody asking for it at six in the morning. |
| Provenance, disclosure and delivery tools | Whether the metadata survives the pipeline, and whether the claim it makes is one a commissioner will accept. |
| Physical capture, on film and digital | The distinctive one. We originate the test material ourselves, the same scene on 35mm and on digital with optics, lighting and camera metadata logged, so a model is evaluated against a measured physical event, not found footage. |
Where a question is outside what we can answer, we say so before a fee is agreed.
Next
Use the Lab, or test something in it.
Productions, vendors and departments can commission a structured test and take the report whether or not it is flattering.
