# google-deepmind-project-genie-3-world-models-2026-02

## Veille

Project Genie - Real-Time Interactive World Models from Google DeepMind

## Titre Article

Project Genie: Interactive World Models with Genie 3

## Date

2026-02

## URL

https://www.youtube.com/watch?v=Ow0W3WlJxRY

## Keywords

World models, Project Genie, Genie 3, Google DeepMind, Google Labs, interactive world generation, real time, Nano Banana Pro, simulation, robotics, embodied intelligence, Simmer, AI agents, interactive media, virtual reality

## Authors

Logan Kilpatrick (Google DeepMind), Diego Schlomi (Google DeepMind), Jack (Google DeepMind)

## Ton

**Profile**: Technical conversation among members of the product/research team, accessible and enthusiastic register, moderate technical level with popularization

**Description**: "Release Notes" podcast format with three speakers from Google DeepMind presenting Project Genie conversationally. The tone is that of a relaxed product demo, alternating between accessible technical explanations and moments of spontaneous discovery during live demos. Enthusiasm is palpable but measured, with honesty about the model's current limitations. The speakers adopt the posture of explorers sharing a frontier technology, inviting the audience to experiment rather than consume a finished product. The target audience is broad: developers, creatives, tech-savvy early adopters.

## Pense-betes

- **World model vs video model**: A video model generates a fixed sequence; a world model generates frame by frame with real-time user control (left/right/actions)
- **Project Genie = consumer app**: The equivalent of what Flow/ISK are for VO, launched for Google US Ultra subscribers
- **Creative pipeline**: Nano Banana Pro first generates a "canvas" image, then Genie 3 turns this image into an explorable world
- **Key personalization**: Uploading personal photos (toys, objects, spaces) to animate them - a differentiator versus other systems
- **60-second limit**: Cost/serving trade-off, dynamism gradually decreases, "two one-minute worlds > one two-minute world"
- **Implicit physics**: The model implicitly learns how the world evolves (balls rolling, water splashing)
- **A harder problem than video**: No freedom to modify past/future frames, must be consistent with both the past AND the immediate action
- **Genie trajectory**: Genie 1 (research paper) → Genie 2 (Dec 2024, 10s, low resolution) → Genie 3 (Aug 2025 announcement, 1 min real-time photorealistic)
- **Simmer use case**: Gemini agent able to interact in 3D worlds, trained in Genie 3 worlds to test generalization
- **Roadmap mentioned**: Multiplayer (complex latency), more controls/interactions, developer API, surfaces beyond the app
- **"Prompting is back"**: The model is not yet robust to prompting, experimentation is required - an opportunity for early adopters
- **Long-term vision**: "A copy of the universe where you can do whatever you want" - a simulation indistinguishable from reality as the north star

## RésuméDe400mots

Google DeepMind is launching Project Genie, a web application that lets Google US Ultra subscribers create and explore interactive worlds generated by the Genie 3 model. Unlike conventional video models that produce fixed sequences, a "world model" generates the environment frame by frame in real time, letting the user navigate and interact.

**Workflow and creative pipeline**: The user starts by describing their world and character. Nano Banana Pro first generates a "canvas" image that serves as a visual starting point. Clicking "Generate World" prompts Genie 3 to turn this 2D image into an explorable 3D environment. The transition from 2D to immersive 3D is the "wow moment" identified by testers. The application also allows uploading personal photos - a photographed toy dinosaur can become a controllable character in a reconstruction of the room.

**Technical challenges**: Genie 3 tackles a more complex problem than standard video generation. A video model can retroactively adjust frames to ensure consistency; Genie 3 must generate in real time, consistent with both the past AND the user's immediate action, without knowing future inputs. The current 60-second limit results from a trade-off: the world's dynamism tends to gradually decrease, and serving costs remain high.

**Evolution since Genie 1**: Genie 1 was a research paper. Genie 2 (December 2024) offered 10 seconds at low resolution, without real time. Genie 3 (announced August 2025, now launched) reaches one minute in real time with photorealistic quality. The team notes that a year ago, a minute of real-time consistency seemed an ambitious goal; today, users are asking for more.

**Envisioned applications**: Beyond entertainment, the team is exploring education (personalized therapeutic exposures, such as a child exploring a room full of virtual spiders) and embodied intelligence. The Simmer project already uses Genie 3 to train AI agents capable of accomplishing goals in arbitrary 3D worlds - a step toward embodied AGI.

**Outlook**: The roadmap includes multiplayer (complex because of latency), more interaction controls, a developer API, and expansion to other surfaces. The team estimates it has reached 50% of its vision, with "enormous headroom" for further improvements. The ultimate vision: a simulation indistinguishable from reality, "a copy of the universe where you can do whatever you want".

## GrapheDeConnaissance

- Google DeepMind —a_créé→ Genie 3 (TECHNOLOGIE, 0.99)
- Google DeepMind —publie→ Project Genie (TECHNOLOGIE, 0.99)
- Project Genie —est_basé_sur→ Genie 3 (TECHNOLOGIE, 0.98)
- Project Genie —utilise→ Nano Banana Pro (TECHNOLOGIE, 0.97)
- Genie 3 —remplace→ Genie 2 (TECHNOLOGIE, 0.98)
- Google DeepMind —publie→ Genie 2 (TECHNOLOGIE, 0.95)
- Genie 3 —est_instance_de→ modèles de monde interactifs (CONCEPT, 0.95)
- Google DeepMind —affirme_que→ les world models résolvent un problème plus difficile que les modèles vidéo (AFFIRMATION, 0.97)
- Simmer —utilise→ Genie 3 (TECHNOLOGIE, 0.96)
- Simmer —fait_partie_de→ Google DeepMind (ORGANISATION, 0.9)
- Google Labs —collabore_avec→ Google DeepMind (ORGANISATION, 0.97)
- Genie 3 —s_applique_à→ intelligence incarnée (CONCEPT, 0.88)
- Logan Kilpatrick —dirige→ Release Notes (EVENEMENT, 0.99)
- modèles de monde interactifs —permet→ génération temps réel (CONCEPT, 0.96)
- Google DeepMind —prédit→ simulation indistinguable de la réalité (AFFIRMATION, 0.82)

---
Canonical: https://www.thekb.eu/en/fiches/google-deepmind-project-genie-3-world-models-2026-02/