Generative AI dominates the conversation, but faster and more reliable television production does not come from offloading creative decisions to a large language model. It comes from understanding context.

This session demonstrates how Limecraft combines camera-to-cloud workflows with production metadata such as scripts, script notes, continuity data, time-codes and automated transcripts. By connecting these sources while filming is still under way, the platform can automatically organise rushes, synchronise material, identify relevant takes and generate reliable pre-cuts for editorial review.

The approach is grounded in real production environments, including entertainment formats like The Voice (ITV Studios) and continuing drama as produced by Warner Bros. These cases show that effective automation depends less on randomly applying AI than on capturing, structuring and interpreting the metadata created throughout production. AI is used selectively, where it adds measurable value, while deterministic workflow logic provides accuracy, traceability and control.

The objective is not to automate storytelling or generate the final pixels. It is to remove repetitive technical work so editors and creatives can concentrate on narrative, performance and editorial judgement. Bringing editorial preparation much closer to the moment of production also means that decisions can be made while the material and its context are still fresh in people’s minds.

The result is a shorter path from set to post, faster feedback during production and substantial consequential savings in editorial capacity, without compromising creative control.

In this session, we will explore in detail the anatomy of successful automation.

Generative AI is usually shown through finished images and videos. In the daily work of a motion designer and AI filmmaker, the more interesting changes happen much earlier, when an idea still has to be understood, tested and agreed on.

Sylvia Nitzsche walks through four production cases from her own work: a corporate film where AI turned a detailed text script into something the client could actually see and discuss before production; an advertising project where a rough AI prototype tested whether the idea worked before more time and budget went into it; a music video where building a consistent avatar raised very human questions about consent, self-image and trust; and a virtual presenter that had to speak and gesture for several minutes and hold up live on stage.

Not a prompt tutorial and not a workflow to copy, but an account of what happened around the tools: what made the projects faster, what created new problems, where AI helped communication with clients, and what she would decide differently today.

Voia’s cloud-based platform replaces LED volumes, green screens and specialised tracking rigs with real-time AI compositing: 4K video matting with hair-level detail, subject relighting, 3D camera tracking and instant placement into Unreal Engine environments.

That puts virtual production within reach of an entirely new audience. For established studios it means faster previsualisation and more flexible, less infrastructure-heavy workflows that open up new creative possibilities for higher-end productions.

Avner Braverman and Daniel Kovacs discuss how Voia is closing that gap from both directions at once, who gets to make cinematic work, what ‚cinematic‘ requires in the first place, and what it all means for the future of virtual production as AI keeps shrinking the distance between an idea and a finished shot.

Footagen turns an already written script into a finished film. The script becomes one source of truth: it is broken into scenes, shots, characters, locations and props, and that structure carries through planning and generation shot by shot. Several models are orchestrated rather than one being bet on, and every shot keeps a record of the model, prompt and references behind it. The point is not the single pretty frame but getting a whole film out the other end with the director still making the calls.

Jukka Karhula walks through the tool and explains the benefits: what it saved, where it guessed wrong, what had to be regenerated, and how the human still has to always keep the control of the output.

Built in Helsinki and stress-tested inside Frame Rate’s own team, with feedback from filmmakers outside. Betatesters are welcome and can sign in at footagen.ai

When a five-person team from Poland entered the AI Film World Championship, the hard part was not the tools. It was people: who does what when anyone can suddenly do almost anything, and how a team keeps talking when the old job titles stop making sense.

Poland won the Cup, and the team’s film ‚Elegy‘ placed third among 510 entries from 16 countries. Between rounds the roles were redefined every time, and that turned out to be the most valuable skill of all.

Maciej Zemojcin shares what the competition taught him about using AI to streamline communication between people, and about the new roles emerging across film and media. Everything comes from practice, from a contest where more than a thousand people were working it out at once, including how the work was split and re-split each round, what broke and what made the team faster.

One micro-drama, run through a complete AI pipeline live on stage: writing the hooks, locking character consistency, generating shots, directing the AI, dubbing into multiple languages and cutting the final vertical.

Aenn-Sophie Siebert and Jordan Ivey of Hinterhof Verticals show the real process rather than the finished result, including the real failures and the real craft decisions behind creating with AI.

Using a real production project as an example, Robert Förster shows how AI makes existing audiovisual content usable for new markets. The Radiozeit Workflow Engine recognises speakers and music, separates audio tracks, translates dialogue and generates localised language versions.

At the same time, music recognition helps identify licence-relevant content, check rights and, where needed, replace music with suitable tracks from cleared catalogues. Older episodes can be distributed after automated upscaling, and a Content Composer creates editable drafts for new target audiences and platforms, while editorial control and final approval remain in human hands.

What happens when AI moves beyond individual prompts and becomes part of the entire filmmaking process? This session takes a film from initial idea to production inside one tool (MINT), showing how an AI-native pipeline connects the different stages of filmmaking and how AI agents can join the process as creative collaborators rather than just generation tools.

To wrap things up, we’ll go off-script with two recent AI video experiments: enhancing real footage with generative AI and live video generation for interactive movie experiences. How far can you push AI video before things get out of hand? We tried.

Every AI conversation in media eventually arrives at the polished cut or the published story. This session starts one step earlier, at the raw feed, the moment content first exists as audio or video.

Trint founder Jeff Kofman shows how live ingest turns a stream into speaker-attributed text within seconds of something being said, and how automatic metadata extraction pulls out the names, time codes, topics and key soundbites that would otherwise cost an assistant producer an hour.

Rather than one tool in isolation, the session follows a single continuous chain: a camera capturing in the field, a live signal carried back over a bonded connection, transcription and tagging as the feed arrives, a newsroom system pulling the tagged transcript into a rundown, and an editing suite cutting from that same searchable text. Including the places where a human has deliberately been kept in the loop instead of the judgement calls being automated.

In this 30-minute AI Garden session, ART+COM demonstrates how its 3D generation engine creates 3D spaces and objects from simple inputs such as text, photos, videos or speech.

Step by step, the session covers generating and refining 3D environments in real time, explores concrete use cases ranging from investigative storytelling to spatial prototyping, and shows how iterative, consistent scene creation works. A hands-on look at the current state of AI-driven 3D creation.

Presented by GenKI4Media, a project dedicated to the development of generative AI assistants for the media, cultural and creative sectors and funded by the Federal Ministry of Research, Technology and Space.

Can a Hollywood-grade pipeline run on indie hardware? Over the past few years Alican Eren Kuzu set out to replicate sophisticated studio workflows with low-cost equipment and software, with the aim of making high-end animation and CG-heavy films possible on European budgets or less.

The session covers the setup he used to shoot, build and rough-cut a 105-minute animated feature solo in Unreal Engine 5, at home and in his free time: performance capture with pose-estimation AI in low-cost spaces such as an office boardroom, a sports hall or a living room; fully AI-generated animation edited to fit; and sets, assets and cameras linked to master structures, so a single change propagates across hundreds of scenes without manual rework. The entire film lives in one timeline of 199 scenes, editable on the fly.

The resulting previs carries camera, depth and other passes and can feed video-to-video models, and the setup scales from a solo artist to a full crew, serving as a foundation for feature previs, LED volume prep or full CG animation.