AI 热点

Google 向符合条件的 AI Ultra 用户开放 Project Genie 中的 Genie 4

Unite.AI更新于 11月19日 08:53

Google Brings Genie 4 to Project Genie for Eligible AI Ultra Users

AI 导读

Google 向符合条件的 AI Ultra 用户开放 Project Genie 中的 Genie 4,Google AI Pro 用户可加入候补名单,但不在初始开放范围内。Genie 4 支持用户在探索过程中通过提示词改变天气、角色、物体、镜头和事件,并同步生成环境声、音乐及角色对话。Google 指出其长时间运行时仍可能出现视觉细节漂移、物体持久性下降、镜头延迟及空间精度问题。

推荐理由

AI 评分 74

文中并列呈现 Genie 4 的会话内控制与原生音频能力,以及长时一致性和空间精度等限制,便于评估其创作与训练用途。

主题

标签

正文

Google AI Pro members can join a waitlist on the Project Genie site, but are not part of the initial rollout.

Genie 4 adds control during a session. Users can type prompts while navigating to alter weather, add a character or object, move the camera or redirect events without starting over. The model also generates native audio with the visuals, including ambient sound, music and character dialogue that reacts to user inputs and events in the simulation.

Gemini’s Role in World Simulation

Google DeepMind describes Genie 4 as natively multimodal, with Gemini Omni providing a real-world reference for how simulations look and behave and other Gemini models supplying reasoning for coherence and continuity. Gemini 3.8 Flash can serve as an in-loop guiding agent: it observes the scene, splits goals into steps, checks incoming frames for progress and changes its plan when the world or user input shifts.

Google DeepMind presents this coupling as a way to model multi-step events and test how AI systems navigate environments before real-world deployment. Intended uses include training environments for robots and agents as well as new interactive experiences.

Creation, Sharing and Model Choices

The Project Genie agent handles initial creation through natural-language conversation. It coordinates Google’s multimodal models to produce the world’s opening visuals and a custom background score suited to the intended mood. During exploration, Gemini can watch the screen, suggest context-aware next actions through on-screen chips and trigger events automatically.

The gallery adds a publishing and reuse layer. Users can alter an existing world’s structure, soundtrack or rules with the agent, then publish their version for others to explore and remix. Examples range from Living Memory, which converts personal photographs into explorable spaces, to The Life of Ants, created with filmmaker Anand Gandhi, which presents a colony from an ant’s perspective.

Google positions Genie 4 as its higher-fidelity option for visuals, physics and native audio, while Genie 3 remains selectable for longer runs or worlds seeded from 360-degree panorama images. Genie 3’s per-generation limit is now three minutes, up from 60 seconds, and users can link multiple worlds into longer sequences or trigger real-time events with natural-language prompts.

Safety and Unresolved Limits

All video and audio generated in Project Genie include SynthID, Google’s imperceptible watermark for identifying AI-generated content, and all model inputs and outputs undergo safety screenings.

Google says both Genie models remain frontier research systems and identifies camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence as unresolved challenges. Visual details and object permanence in Genie 4 can drift during longer sessions, after complex edits or while navigating; legible text and spatial accuracy also remain active research areas.

The launch expands who can build with Genie 4, but the durability of its simulated worlds remains the key constraint on longer creative sessions and on Google DeepMind’s proposed training environments for robots and agents.

来源

Unite.AI