The 2026 edition took place on 6 October 2026 in the cinema of the National
Film School of Denmark in Copenhagen. It was curated by Sten-Kristian
Saluveer, CEO of Storytek Innovation Lab, Curator of Cannes Next at the
Marché du Film and Head of Studies of Mastering AI for European Film
Professionals, and hosted and moderated by Wendy Mitchell, contributing
editor at Screen International and host of The Screen Podcast. Saluveer, called
up for reservist duty in Estonia, opened the day with a recorded address.

Twelve speakers took the stage across six sessions, with two screenings and an
evening reception. The opening fireside chat was held under the Chatham
House Rule, so this public report summarizes its themes and recommendations
without attributing them to individual speakers; the other sessions are reported
with their speakers named. Picture This 2026 was supported by Nordisk Film
Fonden, PRD Kulturplus, the Danish Directors Guild and FAF.

Capture is an asset. A set that is dressed, lit and scanned once, with the
lighting baked in, can be extended, reframed and reused by compositors
long after it has been struck, which frees the camera and shortens time on
the most expensive part of a shoot. The quality of the dataset decides
whether that asset survives the next generation of tools, so the
recommendation to crews is to acquire good photo-based datasets of sets
and props now.

Adoption follows allies. New tools reached the production floor when a
head of department, typically the VFX supervisor or the cinematographer,
took the decision and carried it to the director. Demos and one completed
use case persuaded colleagues more than any argument, and R&D should
report its failures alongside its successes.

Beware the android trap. When the machine generates and the human
merely approves, the result is technically competent but emotionally empty,
and skills atrophy, above all in the entry-level jobs where people learn by
doing. The alternative keeps the human leading and the machine amplifying:
prompt it, challenge it, dance with it, but do not delegate to it.

Talk about cost avoidance, not only cost cutting. The more useful question
is what a production can now afford to attempt, from digital animals to
effects previously beyond the budget, rather than what it can remove.
Judgment is the durable skill. The tools will look dated within a couple of
years; what lasts is knowing where and when to use them, when not to build
one, and how to keep the craft alive. Practitioners learn what comes next
from their community, not from a search engine.

Humans stayed in the loop at every step. Writer-director Tiffany Lin built the characters from museum research and her own drawings and approved each stage from hand-drawn animatic to post. ‘The idea of prompt and pray is not something we’re looking for,’ Ivy said, adding that tracking creative intent is crucial to securing copyright.

Generative output is not cinema. ‘AI did not create cinema. AI created
images and video.’ A VFX supervisor’s compositing recipe, a traditional
color pass at FotoKem and conventional sound design were needed to put
the fish in the deep, and the arrival of a new model late in production meant
redoing about 80% of the images in three weeks.

The biggest pain points were human: film-trained crew and AI artists used
the same words to mean different things, and an AI artist’s skill is hard to
judge beyond a reel.

Copyright decides the feature question. Studios will not invest until they
control the primary IP, because ‘if you can’t copyright Nemo you don’t have
lunchboxes, and kidswear, and toys’.

‘Post is now part of pre-production.’ Editors are increasingly on set, with
comps edited during filming to avoid costly reshoots. Development and
storyboarding are the easiest entry points for filmmakers.

Virtual production is an ecosystem of wall, camera, lighting and
environments, kept open so filmmakers can work with the cameras and
lights they already know. ‘We can’t expect people to come to us,’ he said;
productions follow incentives, crews and locations, so the stage must
travel.

Much of it needs no wall. Previsualization served purely as a planning tool
on I Play Rocky, and Gaussian splatting works as ‘reshoot insurance’: an
hour’s scan of a location can be brought onto a stage later.

The savings are ‘a pencil sharpening exercise’, concentrated in locations,
visual effects and post, with roughly 80% of a VFX shot captured in camera.
The trade-off is that decisions move into pre-production.

Training is central. Film schools, unions and film commissions ask first
whether local crews can be trained; the shortage is trained people, not
hardware. Film schools should teach virtual production the way they should
teach distribution: well enough to judge when to use it.

‘It’s still just a magic trick.’ Roles change rather than disappear, and an LED wall is always lit as though it were a real location.

Imagination is humanity’s only chance of survival. We live in a crisis of
fiction, in which true stories are valued above invented ones, while
politicians treat the future as fixed; imagination ‘shows the possibility of the
alternative’.

Reality is her raw material: ‘I like the naked things,’ she said, shaping them
into narrative afterward. Franz is a puzzle rather than a biopic, a search for
Kafka under the interpretations and kitsch.

Trust is practically the only thing an actor needs from a director; actors ‘just
have to feel that they are safe with you’. The cinematographer is her closest
collaborator and a producer she can trust is a gift.

Green Border was shot in 24 days once the border zone was closed to
media and ‘my tool is fiction’ became the only way in; a big festival offered
protection and visibility.

On technology she is open to everything, with one condition: ‘I need to have
the final control.’ AI’s danger lies in news and politics, not fiction, and AI
revoicing that keeps an actor’s own voice is better than dubbing by
someone else.

European cinema’s problem is not money, technology or audiences: ‘The
problem is that our movies are too mediocre.’ Filmmakers must free
themselves from the comfort of support systems, festival selections and coproductionsand find the courage to make personal films.

Tiller calls himself ‘not a director, but a prompteur’ and the work a
promptus rather than a film. Its protagonist, Loab, a figure that emerged
from a negative prompt, reveals for Tiller a great deal about what sits in a
model’s training data.

‘You don’t need light anymore to make a film.’ The model reduces statistical
noise to an image rather than capturing reflected light, and the anxieties
that greeted the birth of cinema are now projected onto AI.

Control came slowly. He began with DALL-E in late 2022 and only in early
2025 had enough control; ‘Photoshop was the lifesaver’, references and film
stills worked better than negative prompting, and sound and editing were
done the traditional way. When a mirror shot came back with unplanned
reflections, ‘the models become a collaborator of sorts’.

The prompter has become a third persona beside producer and director,
but the work is too laborious for anyone hoping to prompt and walk away.
AI companies should be made to fund culture, and on copyright, now ‘like a
joke’, regulation is needed; Denmark’s Ministry of Culture proposed a
likeness law the same day.

‘Don’t do AI short films if you want to make a lot of money.’ His next feature
is being developed with the August Sander Archive.

An AI animal can be directed. Madsen trained models only on footage shot
for the film and rendered up to 100 wolf takes a night for Álvarez Mesén:
‘It’s very much like shooting on set with actors.’ Of 5,241 takes, today’s
tools would need about a tenth, and his share of the project consumed
about 5,000 kroner of power.

A bog on a bog. For The Chestnut Man, Buus and Trier Mørk built a 34,000-
liter tank on Nordisk Film’s Shortcut stage and recorded 8K spherical plates
in the real bog; because of the cost in Denmark, such a decision needs the
whole team aligned.

On Høst, 365 VFX shots were done traditionally where possible and with AI
where it was better or cheaper, from rebuilt corn to a relit scene. ‘Generally
I’m not pro AI humans, but in this case I’m very pro not having babies on
set,’ Madsen said. Nothing was impossible without AI, but much had been
out of reach on the budgets available.

The panel split on labor. Zandvliet warned that producers will choose the
cheapest option and ‘the cheapest version is AI’; Buus sees AI as a
companion; Madsen wants it as a recovery tool used after creative
decisions are made on set, and replied: ‘I’m building the tools to enable new
stories.’

Consent and provenance. The team works in a closed system with nothing
uploaded, so the source of every frame can be shown. Windelin urged
producers to write AI use into actors’ contracts, and the industry still has to
define levels of consent.