The Feral Engine

01 — Concept

A situated intelligence.

Not generative AI in the commodity sense, but a situated artificial intelligence dedicated to real-time audiovisual production. The Feral Engine generates operational instructions and a parallel stream of imagery from live capture and a dataset specific to one place — not a model trained on scraped archives, but an authoring engine for the next generation of cinema, immersive work and live performance.

The Engine is a situated system — it needs to be anchored in a real, instrumented environment to learn, sense and operate. Where today's generative models hallucinate from a flattened past, the Feral Engine perceives, infers and operates within a continuous now: capture, edge inference, output, loop.

What it produces is, first, operational instructions — directions to cameras, to sound systems, to lighting, to performers, to entire production crews. The Engine orchestrates the act of filming. It does not replace creation; it amplifies and re-conditions it.

But it does not stop there. Because the dataset is built from a single place sensed continuously over time, the same Engine can also generate images that sit in seamless continuity with what is being captured — a parallel imagery of what is and what could also be. On set these generated images run in low resolution and serve as reference and direction; in post-production and distribution they can be rendered at full resolution and woven into the final film alongside the captured footage. The break between recorded and generated dissolves: same place, same dataset, same authorship.

A new category of cinema — what is, and what could be, authored on the same continuous ground. A cinema of the possible.

Three things, kept distinct

The Feral site · the laboratory

Fifteen hectares of high plateau in central France. The instrumented ground where the Engine is built, experimented and validated. Not the product — the laboratory.

The Feral Engine · the product

A transferable system, designed from day one to be lifted out of The Feral site and re-anchored elsewhere — at a studio, in a museum, on a film set. The Feral site provides the conditions for development; the Engine itself is not bound to it.

The Feral Artworks · the proof

A long-arc artistic project on The Feral site: successive Epochs, each authored by a major international artist over a defined window. Epoch 1, by Fabien Giraud and the founding team, is in production; Pierre Huyghe is announced as the lead artist of Epoch 2. Each Epoch is at once an authored artwork and a full-scale deployment case for the Engine.

The next decade of AI in the cultural industries will not be won by ever-larger models trained on scraped archives, but by controlled, situated, sensor-rich systems that can be deployed, audited and trusted by serious creators and institutions.

The Feral Engine is built from the ground up for that world — at a moment when the cultural industries face a generative AI shock with no infrastructure adapted to their workflows, rights regimes or aesthetic ambition, and when native IP traceability is becoming a regulatory requirement (EU AI Act, Content Authenticity Initiative).

02 — Technology

One continuous pipeline.

The Engine is structured around a seven-block architecture that turns raw environmental signal into authored real-time output.

Two regimes — continuous learning (off-shoot) and situated activation (on-shoot) — coupled through the latent space at Block 4. The system perceives, learns, generates, activates and transforms the real, which loops back as new perception.

Close-up · the activation chain · unfolding of blocks 5–6

A single beat of the activation loop, expanded. The Engine captures one take, generates six alternative versions in parallel (image + sound), and presents them to the director and authors. The chosen take becomes the new ground truth — and is translated back into revised instructions to humans and machines, producing a new state of the real. The cycle repeats, take after take, until the scene is closed.

Architecture

01 Capture

Multi-modal sensing across 15 ha — image, sound, telemetry

Cadmos

02 Edge

On-site low-latency inference and signal routing

Cadmos

03 Orchestration

Cloud / hybrid pipeline, scheduling, model registry

DEV-ID · OREUS · DataGreen

04 Curious AI

Site-trained models that learn the environment over time

Inria Bordeaux

05 Synthesis

Real-time image and sound generation and composition

CNRS · XLIM

06 Traceability

Native provenance chain and IP attribution

Alien Intelligence

07 Activation

On-site stage for validation, demonstration, audiences

The Feral

Situated grounding

Anchored in a specific instrumented environment. Radically lowers hallucination, makes outputs auditable, and gives every generation a verifiable spatio-temporal context. For institutional customers — broadcasters, museums, regulated industries — this is the difference between an interesting demo and a deployable system.

Native traceability

The IP and provenance layer is not a bolt-on. Every operational instruction carries a signed lineage of inputs, models and authorship. Aligned with the EU AI Act and the Content Authenticity Initiative from day one.

Transferability

The R&D phase deliberately targets a system that can be lifted out of The Feral and deployed elsewhere — at a studio, in a museum, on a film set. TRL 4 today, TRL 7 by 2029. The technical core of the commercial model.

Validation at The Feral, by international artists

Validation happens at The Feral site itself, through a long-arc artistic project structured as successive Epochs — each authored by a major artist over a defined window. The Engine is built, exercised and stress-tested by the work of these artists at The Feral site. Epoch 1, by Fabien Giraud and the founding team, is currently in production; Pierre Huyghe is announced as the lead artist of Epoch 2. Each Epoch is at once an authored artwork and a deployment case study for the system — the works are not by-products of the technology, they are how the technology proves itself.

Recent international presentations: The Feral, Epoch 1 at Berlin Atonal (Kraftwerk, May 2026); Strange Rules curated by Hans Ulrich Obrist at Palazzo Diedo, Venice (Berggruen Arts & Culture, April–November 2026); NY AI & Creativity Summit (May 2026).

03 — Team

A coalition of partners.

An unusual alliance — working artists, senior research laboratories, and dedicated technology partners assembled around a single project. It is carried by Association 3024, a non-profit lead organisation, in consortium with Alien Intelligence, Fair Decision, the CNRS and Inria, and structured around three complementary poles — the Institute (innovation), the Collective Work (validation in real conditions), and the Studio (transferability and deployment) — steered by three committees: scientific, artistic and transferability.

Direction · click a name to expand

Fabien Giraud

General Director · Artistic Co-direction

French artist and author whose work, rooted in documentary practices, explores the hypothesis of other possible worlds through sculpture, film, and performance. From 2014 to 2022, he developed The Unmanned with Raphaël Siboni — a polyphonic, inverted historical epic shown at MONA, Liverpool Biennial, Okayama Triennale, Biennale de Lyon and Palais de Tokyo. In 2023, he initiated The Feral with Anne Stenne. Within the project, he leads conception and orchestration: defining the artistic orientations and human-AI co-creation protocols, designing system use under real production conditions, and steering the development of the device as an operational tool for the cultural and creative industries.

Anne Stenne

Artistic Co-direction · Director of Projects & Validation

Independent producer and curator specialised in the conception and execution of works with complex protocols across institutional and international settings. She develops long-arc collaborations with artists such as Pierre Huyghe and Fabien Giraud, and has led curatorial projects for first-rank institutions (Fondation Beyeler, LEEUM Museum of Art, Pinault Collection). She curated Infantia (1894–7231) at IAC Villeurbanne and edited the catalogue The Unmanned (Mousse Publishing, 2022). Within the project, she co-directs artistic strategy and runs the validation pole: designing curatorial protocols and artist intervention frameworks, organising the conditions under which the system is tested in real environments (Epochs), and interfacing with cultural institutions and international partners for validation and diffusion.

Ida Soulard

Director of Training & Transmission Protocols

Art historian, curator and researcher, PhD from École normale supérieure (SACRe / PSL programme). Her work sits at the intersection of contemporary art, theories of milieu, and situated forms of knowledge — with particular attention to the relations between artistic, technical, and environmental practices. She directed the Fieldwork: Marfa international research and residency programme (2013–2020) and co-founded the research platform Glass Bead. She pursues a curatorial, editorial, and pedagogical activity articulating knowledge production, artistic experimentation, and transmission formats (ENSA Bourges, publications, seminars). Within the project, she co-directs the Institute and leads training: pedagogical protocols, situated formats (residencies, seminars), and the theoretical frameworks accompanying new uses of AI in artistic practice.

Grégory Chatonsky

Co-director of Innovation

French-Canadian artist and researcher whose work focuses on the relations between technical systems, memory, and generative processes. He has developed for over twenty years a practice at the intersection of digital creation, media theory, and artificial intelligence. Founder of the collective Incident.net in 1994, he began pioneering work on generative systems and artistic uses of AI in close dialogue with academic and public institutions (ENS, French Ministry of Culture, contributions linked to the EU AI Act). He teaches at Le Fresnoy, UQAM, and Musashino Art University; his work has been shown at Centre Pompidou, Palais de Tokyo, and Jeu de Paume. Within the project, he contributes to the design of generative and agentic AI architectures — modelling generation processes, system-environment interaction, and new regimes of co-writing with AI.

Martial Geoffre-Rouland

Co-director of Innovation

Creative technologist and designer of interactive systems, based in Marseille. Trained in interactive design at Elisava (Barcelona), he works at the intersection of machine learning, computational design, and interactive environments, with strong experience deploying experimental devices under real conditions. Founder of Screen Club, he designs and deploys bespoke architectures combining generative AI, computer vision, and interactive systems for artistic, cultural, and technological contexts. He has led training and structuring activities for AI use at Centre Pompidou and MIT (in collaboration with Google Arts & Culture), and collaborates with major artists, notably Pierre Huyghe. Within the project, he contributes to development and integration of the generative and agentic AI architectures — prototyping, real-time interaction, and the deployment of hybrid capture / processing / generation devices.

Florence Cohen

Director of Production

Producer specialised in audiovisual and cinematographic projects with a strong artistic dimension. At Anna Lena Films, she has accompanied over the past decade the production of films by international artists and directors — among them Pierre Huyghe, Philippe Parreno, Dominique Gonzalez-Foerster, Jesper Just, Sophia Al-Maria, Marine Hugonnier, and Tomás Saraceno. She handles financial structuring, production coordination, and project oversight in demanding contexts at the interface of cinema, contemporary art, and technology. Within the project, she leads production: budget structuring, resource piloting, partner coordination, and the operational conditions enabling the device's deployment in real situations.

Jean-Christophe Radke

Director of Technical Operations

Head technician specialised in the conception and implementation of technical devices for artistic projects and exhibitions of high operational complexity. His career combines deep technical expertise, production coordination, and project management in demanding contexts at the intersection of contemporary art and institutional environments. Through numerous collaborations with artists and public and private institutions, he has developed a practice attentive to the material conditions of artwork production, articulating reflection on devices with mastery of technical constraints. He has notably worked with Katharina Grosse, Maurizio Cattelan, Christian Boltanski, Paul McCarthy, and Jannis Kounellis, and currently holds a position at Collection Lambert in Avignon. Within The Feral, he directs technical operations: coordinating production devices, implementing capture infrastructures, and articulating artistic requirements with technical constraints.

Bureau · Association 3024

François Quintin

President · Director, Collection Lambert (Avignon)

Jules Rimbaud

Vice-President & Treasurer · Director US, Excurio

Nathalie Viot

Secretary · Curator & cultural projects director

Artistic committee · steering

Hans Ulrich Obrist (Serpentine Gallery, London) · Noam Segal (Guggenheim Museum, New York) · Suzanne Cotter (Museum of Contemporary Art, Sydney) · Anna Longo (philosopher of technique) · Anna Lena Vaney (producer) · Camille Bréchignac · Sandra Terdjman.

Consortium · industrial & research partners · click a partner to expand

Alien Intelligence Primavera De Filippi & Léo Blondel

IP traceability · rights management for generative AI

French company specialised in tracing AI-generated content and managing authors' rights in the era of generative models. Mobilised team: Primavera De Filippi (Head of Legal Innovation) — CNRS Research Director and Faculty Associate at Harvard's Berkman Klein Center, author of Blockchain and the Law (Harvard University Press) and Blockchain Governance (MIT Press); Léo Blondel (Scientific & Technological Director) — co-founder of Alien, PhD in computational biology from Harvard, co-founder of Just One Giant Lab; Ghislain Delabie (Solutions Architect) — 18+ years in AI architecture, data governance and applied innovation. Within The Feral, Alien Intelligence provides the trust infrastructure: end-to-end data orchestration, prompt-level tracing of every contribution (datasets, styles, LoRA models) regardless of how content is recomposed by generative AI, and automated rights attribution that compresses today's 3-9-month redistribution delays. Their work transforms cultural data into a traceable, governed, monetisable asset — without substituting for existing collective-management organisations, but augmenting them technically.

Fair Decision Michalis Lianos & Alexander Polonsky

Participative AI governance · quintuple-helix deliberation

French non-profit dedicated to strengthening democratic participation within institutions and territories. The team develops a socio-computational platform for direct, iterative, AI-assisted democracy — adapted to complex multi-stakeholder environments. Founded by Michalis Lianos (professor of political sociology, PhD in law and society, author of several books on direct democracy, social norms, and political behaviour) and Alexander Polonsky (data scientist with 20+ years in semantic analysis and AI-assisted decision-making, PhD in neuroscience). Their methodology — the "black box" framework — allows non-experts to contribute to complex decisions by reacting to expert projections, structured through a quintuple-helix model articulating civil society, public institutions, research, private sector, and creative community. Within The Feral, Fair Decision deploys the participative AI-governance layer, making the project not only a cultural and technological infrastructure but a co-construction device with communities, artists, scientific partners, and local actors of the territory.

Inria Bordeaux Flowers team · Pierre-Yves Oudeyer

Agentic AI · curiosity · autonomy in real environments

France's national research institute in digital sciences and technologies — internationally recognised in AI, robotics, machine learning, and complex systems. The project draws principally on the Flowers AI & CogSci team (Inria Bordeaux Sud-Ouest), specialised in developmental learning, artificial curiosity, and autonomous systems. Led by Pierre-Yves Oudeyer (Inria Research Director, head of Flowers), an international reference on intrinsic-motivation learning and autotelic systems, whose work over twenty years has modelled curiosity in humans and machines — particularly an agent's ability to set its own goals to explore an open world. The team mobilises an interdisciplinary collective of researchers, engineers and PhD students in reinforcement learning, autotelic agents, human-machine interaction, and learning dynamics. Within The Feral, Inria transforms the prototype into a situated, agentic, developmental AI: an autotelic engine that generates its own operational instructions, a language model grounded in the site, and co-evolution mechanisms between the curiosity of the AI and that of the humans involved.

CNRS · XLIM Axe ASALI · Larabi / Crespin / Sanchiz-Viel

Computer vision · multimodal capture · TRL validation

Multidisciplinary research institute (UMR CNRS 7252) attached to the universities of Limoges and Poitiers — recognised for its work in image processing, computer vision, and modelling of complex visual environments. The project draws on the ASALI axis (Image Synthesis & Analysis), led by Mohamed-Chaker Larabi (Professor at Université de Poitiers, ASALI head), with Benoît Crespin (Professor of Computer Science at Limoges, head of the SIR team — Simulation, Image, Rendering) and Nathan Sanchiz-Viel (Senior Lecturer at XLIM, specialist in computer vision and 3D data processing: point clouds, laser scanning, structuring of spatial data from capture). Around this core, XLIM mobilises an extended collective of doctoral students and research engineers — deep learning for images, photogrammetry and 3D reconstruction, multispectral compression, model frugality. Within The Feral, XLIM contributes reliable acquisition protocols (potentially multispectral), temporal and multi-sensor datasets, image-enhancement algorithms (super-resolution, dehazing, deraining), 3D reconstruction by learning, and semantic compression — consolidating the scientific basis required to move the system from TRL 4 to TRL 7.

Technology & service partners · click a partner to expand

Cadmos

Capture & edge · cinematographic equipment

Partner for cinematographic equipment acquisition and audiovisual-flow management between the shooting set and the local AI infrastructure. The Cadmos edge node is deployed within 500 m of capture zones, allowing energy and data flows to the local datacenter to be optimised — minimising latency, transfer costs, and energy losses inherent to long-distance media pipelines.

DEV-ID · OREUS · DataGreen

Edge GPU cluster · sovereign hybrid cloud · pipeline ops

The project relies on a two-tier hybrid architecture. First, a local edge layer with DataGreen: a compact GPU cluster installed as close as possible to the Feral site (1-3 high-density GPU servers, 8-24 GPUs), allowing local data processing, reduced transfers, lowered latency, and a controlled energy footprint. Second, a cloud extension with OREUS / DataGreen for heavier or one-off needs — model training, large-scale inference, distributed storage, and hybrid orchestration. DEV-ID handles operations and application maintenance of the AI pipeline: supervision, updates, and technical evolutions.

Metalaw

Legal architecture · IP & AI Act compliance

The specialist law firm securing the project across intellectual property, data law, digital law and contract law — patents, trademarks, copyright and neighbouring rights, databases, open-source licensing, GDPR, the Data Act, the Data Governance Act, the EU AI Act, cybersecurity, licensing, consortium agreements and the possible incorporation of a joint commercial company. Metalaw also brings rights-management expertise — collective management, direct licensing, and alternative blockchain-based schemes — to organise the exploitation of contents and data, the valorisation of results, and the remuneration of rights-holders and partners.

Tech'IN

Cybersecurity · data governance across consortium

Cybersecurity of the edge and cloud infrastructure — urbanisation and securing of data across consortium members to limit operational risk and ensure trusted multi-party data exchange in an environment where capture, training datasets, and generated outputs must be governed end-to-end.

BE Energethik

Energy autonomy · dynamic AI workload

Designs and deploys the energy system of the Feral Engine: study, sizing, and implementation of a dedicated photovoltaic supply for the site, alongside real-time energy-monitoring tools. Their role is to evaluate and operationalise the AI system's capacity to adapt its processes (capture, computation, storage) to the energy resources available at any given moment — a dynamic-regulation logic at the heart of the project's energetic autonomy.

Les Augures

Responsible AI · sustainable digital strategy

Expertise in responsible strategy and sustainable digital, applied to the cultural sector. Their role across the three-year programme is to frame and pilot a responsible-AI roadmap — sensitising the team, partners, and territory to the socio-environmental issues of AI; assessing the system's footprint to inform production choices; and formalising indicators, eco-design scenarios, and transferable resources for the cultural and creative industries at large.

Correspondances Digitales

Economic engineering · commercial valorisation

Supports Association 3024 and its partners in the review, adjustment, and consolidation of the commercial exploitation plan — commercialisation strategy, validation of business-model hypotheses, activity forecasts, economic impact, job creation, valorisation of the assets developed within the project, and alignment with territorial policies.

International institutional network

Active collaborations with first-rank institutions: Museum of Contemporary Art Sydney, Fondation Berggruen (Palazzo Diedo, Venice), Guggenheim Museum (New York), MAAT (Lisbon), Collection Lambert (Avignon), Jeu de Paume (Paris), Le Fresnoy & Artech–Paris 8 (training); recent exhibitions at Berlin Atonal (Kraftwerk), Lafayette Anticipations.

Site · Plateau de Millevaches

15 ha of R&D land in Cheissoux, Limousin — forests, prairies, peat bogs, streams instrumented with sensors, cameras, microphones and edge servers — plus research-creation studios, residency housing for situated training, and an 800 m² building dedicated to AI-era audiovisual production.

04 — They trust us

An international network of deployment.

A selection of the institutions backing The Feral — through formal letters of support, exhibitions, collaborations, or sustained engagement with the work.

05 — Financing plan

A layered capital stack.

Two phases — a 36-month industrialisation programme (2026–2029) followed by a 36-month scale-up phase (2029–2032) — layered across non-dilutive public funding, project owner co-financing, and a defined private envelope.

Phase 1 · Industrialisation · 2026 – 2029

R&D investment pipeline blocks 1–7

€ 1.41 M

Site infrastructure production stage, edge facility

€ 0.38 M

Operating costs 36 months, four poles

€ 1.12 M

Total Phase 1

€ 2.90 M

Funding mix · Phase 1

Non-dilutive public funding R&D + sovereign real-estate

In instruction

up to € 1.65 M

Project owner co-financing equity, in-kind, patronage

Engaged / in build

€ 0.45 M

Open envelope for private capital

Being structured

€ 0.7 – 1.0 M

Phase 2 · Scale-up · 2029 – 2032

Continued investment

€ 0.60 M

Operating costs

€ 1.65 M

Projected Engine revenue

€ 4.10 M

Recurring institutional support

€ 0.72 M

Phase 1 builds the Engine: half goes into R&D, a third into running the four poles, and the rest into the site & production stage. The funding stack is layered — public non-dilutive carries the deep-R&D risk, owner co-financing covers operations, and a private envelope sits on top with calibrated headroom (€ 3.10 M assembled vs. € 2.90 M needed) to absorb scope flex. By Phase 2, the Engine self-finances: 85 % of the inflow is product revenue.

Where private capital fits. We are not raising a single round. We are designing a layered capital stack — non-dilutive public funding for the deep-R&D risk, strategic patronage for the demonstration works, and a defined private envelope for the company-building of the Engine itself. The investor conversation that this brief is meant to open is on that third layer.

06 — Revenue outlook

Multiple revenue streams.

Revenue is structured around the Engine as a B2B product, with two reinforcing layers — services and artwork — that surround it commercially and aesthetically.

The Feral Engine is the core line, opening commercially in 2029 at TRL 7. The Feral Services and The Feral Artwork activate three years earlier — they generate immediate revenue, prove the system on real projects, and pre-sell customers for the Engine. By 2030 the three operate together.

The Feral Engine · the four offers

Custom deployment

VFX studios, production companies

€ 60 – 100 k / yr

Turnkey deployment

Artist studios, schools, institutions

€ 100 – 200 k / yr

Enterprise licence

Production companies, studios, broadcasters

€ 300 – 450 k / yr

Education licence

Schools, universities, training institutes

€ 360 – 720 k / yr

Each bar shows the annual revenue range per customer for that offer, plotted on a shared k€ scale. Education licences carry the widest range — schools and universities adopt at vastly different scales.

TRL 7 — fully transferable, supportable, documented — by 2029. Commercial deployment at scale from year three.

The Feral Services

Consulting and training on top of the Engine: workflow audits, pipeline integration, programmes for executive teams, operational teams, authors and students. Mission size typically € 10 – 100 k. Lets serious customers buy into the Engine without committing immediately to a full licence, and funds Engine improvements through real customer work.

The Feral Artwork

Films, installations, live performances and immersive works produced on-site. Revenue from patronage, distribution rights, gallery sales, co-production, ticketing and editions. Each major work doubles as a deployment case study for The Feral Engine and The Feral Services.

Trajectory · TRL & revenue ramp · 2026 – 2032

TRL milestones (top) and annual revenue (curve, M€) move together. The 2029 pivot is when the Engine reaches deployable status and revenue accelerates into the scale-up phase.

Cumulative revenue · the pivot from Artwork to Product

Phase 1 cumulative revenue is dominated by The Feral Artwork as the Engine is validated through artworks. Phase 2 nearly doubles the take while the mix flips — The Feral Engine becomes the leading source of revenue.

The Feral Engine The Feral Services The Feral Artwork

By 2032 The Feral Engine reaches structural break-even on operations. The product becomes the dominant share of new revenue; The Feral Artwork continues to generate IP and visibility that is genuinely difficult to replicate. The long-term defensibility is the combination — a working product with a museum-grade artistic practice running on top of it. There is no other team in the market building both: a disruptive technological product built on the greatest artistic experience of its time.

07 — Project timeline

From validation to scale-up.

A 72-month programme structured in two phases — industrialisation (2026–2029) and scale-up (2029–2032) — with seven yearly milestones tracking the Engine from TRL 4 to structural break-even.

A 72-month arc from initial site validation (TRL 4, 2026) to structural break-even (2032). The pivot in 2029 marks TRL 7 — the point at which the Engine becomes a deployable commercial product and Phase 2 begins.