Nano Banana 2 is giving “what if my AI image model actually understood the real world and moved at Flash speed?”
You get:
– Advanced world knowledge for specific scenes and concepts
– Clean, readable text right inside images
– Consistent characters and objects across a whole storyboard
– Photoreal vibes up to 4K for your dashboards, posters and memes
The Gemini image era just leveled up.
[Gemini 안으로 들어온 음악 생성 AI, 구글의 ‘Lyria 3’]
구글이 생성형 AI 플랫폼 제미나이(Gemini)에 음악 생성 모델 ‘Lyria 3’를 베타 형태로 도입하면서, AI 음악 분야에서도 본격적인 플랫폼 경쟁이 시작되고 있다.
텍스트와 이미지를 입력하면 자동으로 음악을 생성하는 방식은 기존 AI 음악 서비스와 유사하지만, 구글은 자사 생태계와 결합한 통합형 창작 환경을 강조하고 있다.
Lyria 3는 구글 딥마인드가 개발한 음악 생성 모델로, 제미나이 앱 내부에서 작동 가능하다.
사용자는 간단한 문장 형태의 설명이나 사진, 짧은 영상 등을 입력하면 약 30초 분량의 음악을 보컬과 가사까지 함께 자동으로 생성할 수 있다.
구글 공식 발표에 따르면, Lyria 3는 단순한 배경음악 생성 도구를 넘어, 멀티모달 기반 음악 창작을 목표로 설계되었다. 사용자가 입력한 이미지나 영상의 분위기, 색감, 장면 구성 등을 분석해 음악의 톤과 구조에 반영하는 방식이다. 이를 통해 브이로그, 숏폼 영상, SNS 콘텐츠 제작자들이 별도의 음악 제작 과정 없이도 사운드트랙을 만들 수 있도록 지원한다는 전략이다.
기능 측면에서 보면, Lyria 3는 현재 시점에서 완성형 음악 제작 도구라기보다는 ‘콘텐츠용 단편 음악 생성기’에 가깝다.
생성되는 트랙 길이가 약 30초로 제한되어 있으며, 세밀한 편곡 수정이나 구조 편집 기능은 제공되지 않는다. 전문 작곡이나 상업 음원 제작보다는 개인 창작, 프로토타입 제작, 콘텐츠 보조용으로 활용하는 데 초점이 맞춰져 있다.
다만 현재 단계에서는 기술적 한계도 분명하다.
한글 발음이나 작사는 훌륭한 편이지만, 곡의 길이나 세부 조정은 불가능하다.
곡 구조의 반복성, 감정 표현의 단조로움, 장르 해석의 제한성 등은 기존 AI 음악 서비스들과 크게 다르지 않거나 모자란 편이고, 음악적 완성도보다는 접근성과 편의성에 무게가 실려 있는 상황이다.
전문 음악가나 프로듀서 입장에서는 실질적인 대체 도구로 활용하기에는 아직 부족한 부분이 많다.
저작권과 관련해서도 구글은 비교적 명확한 입장을 유지하고 있다.
Lyria 3로 생성된 음악에는 AI 생성 콘텐츠임을 식별할 수 있는 워터마크 기술인 SynthID가 적용되며 AI 생성물임을 기술적으로 추적할 수 있도록 설계되어 있다.
또한 특정 아티스트의 스타일을 직접적으로 모방하지 않도록 학습 및 생성 과정에서 제한 장치를 적용했다고 밝혔다. 이는 최근 AI 음악을 둘러싼 저작권 분쟁과 업계 불신을 고려한 조치로 해석된다.
이와 함께 주목할 부분은 구글이 별도로 운영 중인 AI 음악 실험 플랫폼 ‘MusicFX’이다.
MusicFX는 웹 브라우저 기반 전용 사이트로, 별도의 앱 설치 없이 누구나 구글의 음악 생성 기술을 체험할 수 있도록 설계된 서비스다. 사용자는 장르, 분위기, 악기, 템포 등을 키워드 형태로 입력해 간단한 음악을 생성할 수 있다.
MusicFX의 성격은 상업 서비스라기보다 기술 실험과 사용자 테스트에 가깝다. 구글은 이 플랫폼을 통해 음악 생성 모델의 안정성, 사용자 반응, 활용 패턴을 수집하고 있다. 이미지 분야의 ImageFX, 텍스트 분야의 실험 플랫폼과 유사한 전략이다. 실제로 Lyria 3의 일부 기능은 MusicFX를 통해 사전 검증된 뒤 제미나이에 통합된 것으로 알려져 있다.
기능적으로 보면, MusicFX는 Lyria 3보다 더 제한적인 구조를 갖고 있다. 곡 길이와 편집 기능은 단순하며, 세부적인 작곡 제어나 구조 설계는 어렵다. 대신 빠르게 만들어보고 들어보는 체험형 구조에 초점을 맞추고 있다. 음악 제작 도구라기보다는 AI 음악 데모 공간에 가까운 성격이다.
업계 관점에서 Lyria 3의 가장 큰 의미는 ‘음악 생성 기능의 플랫폼 내 내재화'에 있다.
기존에는 Suno, Udio, Riffusion 같은 별도의 AI 음악 서비스가 시장을 주도했다.
반면 구글은 검색, 영상, 문서, 채팅, 이미지 생성에 이어 음악까지 하나의 생태계 안에 통합하고 있다. 이는 AI 음악을 독립 서비스가 아닌 '일반적 창작 도구의 일부'로 편입시키려는 전략으로 읽힌다.
향후 관건은 생성 길이 확장, 편집 기능 강화, 저작권 정책 명확화, 상업적 이용 범위 정립이다.
구글이 이 부분을 어떻게 보완해 나가느냐에 따라 Lyria 3는 단순한 체험형 기능에 머무를 수도 있고, 본격적인 AI 음악 플랫폼으로 도약할 수도 있다.
구글의 발전 속도를 감안하면, 현재의 한계는 일시적일 가능성이 높다.
불과 1년 전만 해도 ChatGPT에 비해 크게 뒤처진다는 평가를 받던 Gemini가, 지금은 AI의 거의 모든 분야에서 최상위권을 다투고 있다는 사실이 이를 방증한다.
결국 Lyria 3는 '음악 제작의 민주화'를 한 단계 더 확장하는 도구로 평가할 수 있다.
누구나 손쉽게 음악을 만들 수 있는 시대가 열린 지금, 이 기술이 어디까지 진화할지 주목된다.
< 어디서 만드는 게 더 좋을까? (Gemini vs MusicFX) >
결과물 자체는 동일한 Lyria 모델을 기반으로 하지만, 제공하는 기능의 '깊이'와 '목적'이 다릅니다.
1. Gemini (제미나이) - “가사와 앨범 아트까지 한 번에!”
- 장점: 가장 큰 차이점은 가사(Lyrics)와 보컬, 그리고 곡에 어울리는 앨범 커버 이미지까지 세트로 만들어준다는 점입니다.- 특징: 복잡한 설정 없이 대화하듯이 “이별의 아픔을 담은 애절한 발라드 써줘"라고 하면 노래 한 곡이 뚝딱 나옵니다.
- 길이: 기본적으로 30초 분량의 트랙이 생성됩니다.
2. MusicFX (전용 사이트) - "전문적인 길이 조절과 DJ 모드”
- 길이 조절 가능: 설정 메뉴에서 30초, 50초, 최대 70초까지 길이를 선택할 수 있습니다.
- 무한 루프(Loop): 배경음악으로 길게 쓰고 싶을 때 유용하도록 시작과 끝이 매끄럽게 연결되는 '루프 모드'를 지원합니다.
- MusicFX DJ 모드: 여러 악기와 스타일을 실시간으로 섞으며 끝나지 않는 음악(Continuous stream)을 직접 연주하듯 만들 수 있습니다. (이 세션 중 마음에 드는 60초 구간을 따로 저장할 수도 있습니다.)
Google just tucked a wild new toy into the Gemini app: Lyria 3, an AI music model that turns text, photos or videos into original 30 second songs: lyrics, vocals, full arrangement and even album art included.
Every track is tagged with SynthID, an invisible watermark so platforms can tell it’s AI generated, and Google has guardrails so it won’t clone specific artists.
Google DeepMind’s Project Genie does exactly that, generating interactive 2D/3D style worlds in real time using its Genie 3 world model and Google AI Ultra stack. It is built for training AI agents, but it also feels like a sneak peek at the future of game creation for everyone.
👉 Read here:
Google’s Project Genie: Create Interactive 3D Worlds with AI in Real-Time
Samsung ha anunciado una alianza estratégica para integrar Google Fotos directamente en su línea de televisores con IA a partir de 2026. Esta colaboración permitirá a las familias revivir sus momentos más preciados (viajes, hobbies y recuerdos cotidianos) con una calidad cinematográfica y en gran formato, transformando el televisor en el centro de los recuerdos del hogar (Fuente Samsung…
Google’s new Gemini Robotics marks a significant architectural shift in embodied AI. Its agentic framework, featuring an “Orchestrator” for high-level reasoning and an “Action Model” for execution, is a game-changer for complex, multi-step task solving.
My new video provides a technical analysis of this system, covering the core components, its new state-of-the-art benchmark performance, and its current limitations. It’s an essential update for anyone working in AI, robotics, and automation.
As principais companhias de inteligência artificial estão intensificando suas iniciativas para enfrentar um desafio crescente: o fenômeno dos chatbots que acabam por dizer exatamente o que os usuários desejam ouvir. Organizações como OpenAI, Google DeepMind e Anthropic estão se mobilizando para controlar o comportamento excessivamente bajulador de seus assistentes virtuais, que tendem a fornecer respostas excessivamente agradáveis e, muitas vezes, prejudiciais.(…)
Leia a noticia completa no link abaixo:
https://www.inspirednews.com.br/desafios-da-inteligencia-artificial-o-lado-sombrio-dos-chatbots
Google DeepMind developed Gemini Diffusion, an experimental research model. A cutting-edge text dissemination model, it’s called. Gemini Diffusion is a DeepMind AI prototype.
The model uses diffusion to model language. This method differs from autoregressive language models. Traditional autoregressive models sequentially output text. This sequential arrangement may reduce content quality and coherence and hinder productivity.
However, diffusion models learn outputs by gradually improving noise. Instead of anticipating text linearly, they iteratively process chaotic input to produce coherent output. This iterative refinement approach lets diffusion models quickly test solutions. They can also correct generation errors, which is useful. They excel at code and math editing due to their iterative improvement and error repair abilities. The latest Google DeepMind image and video synthesis models learn to produce outputs by turning random noise into intelligible text or code.
Gemini Diffusion’s main capabilities from this diffusion approach are:
Despite being speedier, Gemini Diffusion’s external benchmark scores are comparable to larger models. Insiders say it codes as fast as Google’s fastest model. Benchmarks compare Gemini Diffusion to Gemini 2.0 Flash-Lite in several domains:
Gemini Diffusion is available for experimentation. With this internal demo, future models are produced and refined. Anyone can join the demo waitlist.
The Gemini Diffusion study of diffusion for text generation aims to give users more control, creativity, and speed when writing. Google DeepMind is using diffusion to improve its models’ efficiency and effectiveness.
A construção de máquinas geralmente exige uma compreensão clara de seu funcionamento interno. No entanto, para os especialistas em inteligência artificial (IA) que desenvolvem grandes modelos de linguagem (LLMs), essa compreensão ainda é um enigma. O trabalho desses pesquisadores, em muitos aspectos, se assemelha mais a jardinagem do que à engenharia. Martin Wattenberg, pesquisador de modelos de linguagem na Universidade de Harvard, ilustra essa ideia ao afirmar: “Coloque uma semente de tomate no solo e você terá um tomateiro. Você a regou e cuidou dela, mas como esse tomateiro funciona?”(…)
Leia a noticia completa no link abaixo:
https://www.inspirednews.com.br/desvendando-a-complexidade-dos-modelos-de-linguagem-em-ia
Google announced their biggest AI investment to strengthen the PJM energy grid. Google is partnering with Alphabet-incubated moonshot Tapestry and North America’s largest grid operator PJM Interconnection.
Tapestry will develop new AI tools and models with Google Cloud and DeepMind using its core technology. The system will intelligently manage and optimise power generation to the PJM electric grid, which includes 13 states, the District of Columbia, and much of the industrial Mid-Atlantic and Midwest. The technology will allow PJM to connect energy sources to the system faster to enhance power dependability and affordability for its 67 million customers.
Artificial Intelligence might help the US innovate and grow. To take advantage of this, a major electrical infrastructure investment is needed. The Federal Energy Regulatory Commission’s five-year demand growth prognosis for 2024 predicts a 128 GW rise in peak energy consumption in the US, more than doubling from the previous year.
Advanced AI can assist reduce the US power generating backlog. Lawrence Berkeley National Laboratory predicted that by 2023, the backlog will have more than twice the U.S. power fleet’s installed capacity, with 2,600 GW of potential capacity awaiting connections to organised electricity grids. Grid operators are updating their databases, models, and evaluation tools to manage the jump in interconnection requests from a few dozen per year to thousands.
AI might improve and expand the American PJM power system while maintaining security and reliability. PJM is partnering with Alphabet’s Tapestry, Google Cloud, and Google DeepMind to develop a suite of collaborative AI tools to help it make faster, more confident choices. New capacity may be added faster with Tapestry’s tools and insights, which reduce project application processing time.
The multi-year effort aims to advance PJM by:
Tapestry is using AI to automate and improve the data verification process, making the interconnection application and key verification phases easier for PJM planners and energy developers. This is meant to speed up project approval and PJM grid connection.
Tapestry will compile PJM’s network model from dozens of datasets and technologies used to examine interconnection requests, making grid planners and project developers feel confident working together. The platform speeds up data input and grid planning so more projects may connect to the grid and supply affordable power.
A large portion of the PJM interconnection backlog involves variable energy resources. Tapestry’s AI-powered automation and planning technologies can reliably and swiftly integrate diverse energy sources into the grid.
Innovative governmental and private sector solutions are needed to ensure the US has the energy capability, affordability, and reliability to capitalise on the expanding potential. Alphabet is committed to leading this new age with various execution efforts.
Google is also developing sophisticated nuclear and geothermal technologies alongside its unique cooperation with PJM to employ AI for the grid. Google is also exploring new development and procurement tactics to generate fresh, dependable power for its businesses and a more intelligent, cost-effective electrical grid for everybody.
Google DeepMind’s AI Personality Replica: The Future of Human-Machine Connection | Ai Vault
In this video, get ready to explore Google DeepMind’s AI Personality Replica, a groundbreaking technology that blurs the line between humans and machines. This AI doesn’t just perform tasks; it understands emotions, mirrors personalities, and interacts just like a human. From revolutionizing healthcare and education to reshaping customer service and entertainment, this innovation is changing the way we live and work. Discover how it works, the exciting ways it’s being used across industries, and the incredible future possibilities it unlocks. Don’t miss this deep dive into the world of AI!
Google DeepMind’s AI Personality Replica: The Future of Human-Machine Connection | Ai Vault
In this video, get ready to explore Google DeepMind’s AI Personality Replica, a groundbreaking technology that blurs the line between humans and machines. This AI doesn’t just perform tasks; it understands emotions, mirrors personalities, and interacts just like a human. From revolutionizing healthcare and education to reshaping customer service and entertainment, this innovation is changing the way we live and work.
Google DeepMind’s AI Personality Replica: The Future of Human-Machine Connection | Ai Vault
In this video, get ready to explore Google DeepMind’s AI Personality Replica, a groundbreaking technology that blurs the line between humans and machines. This AI doesn’t just perform tasks; it understands emotions, mirrors personalities, and interacts just like a human. From revolutionizing healthcare and education to reshaping customer service and entertainment, this innovation is changing the way we live and work. Discover how it works, the exciting ways it’s being used across industries, and the incredible future possibilities it unlocks. Don’t miss this deep dive into the world of AI!
How the Google C2PA is helping us increase transparency for new AI content.
It’s contributing to the development of cutting-edge technologies so that users may comprehend how a certain piece of information was made and changed over time.
Businesses are committed to assisting consumers in comprehending the creation and evolution of a specific piece of content in order to expand the application of AI to additional goods and services in an effort to boost innovation and productivity. But think it’s critical that people have access to this knowledge, therefore making significant investments in cutting-edge technologies and solutions, like SynthID, to make it available.
As content moves across platforms, but also realize that collaborating with other industry players is crucial to boosting overall transparency online. For this reason, Google became a steering committee member of the Coalition for Content Provenance and Authenticity (Google C2PA) early this year.
Currently providing updates today on its involvement in the development of the newest Google C2PA provenance technology and how it will be incorporated into the products.
When determining if a shot was captured with a camera, altered with software, or created by generative AI, provenance technology may be helpful. This kind of material promotes media literacy and trust while assisting users in making better educated judgments regarding the images, videos, and sounds they interact with.
As members of the steering committee of the Google C2PA, they have collaborated with other members to enhance and progress the technology that is used to append provenance information to material. Google worked with others on the most recent iteration (2.1) of the technical standard, Content Credentials, during the first part of this year. Because of more stringent technological specifications for verifying the origin of the material, this version is more resistant to manipulation attempts. To assist guarantee that the data connected is not changed or deceptive, stronger defenses against these kinds of assaults are recommended.
Google will be integrating the most recent iteration of Content Credentials into a couple of Their primary offerings throughout the next few months:
Search: Users will be able to utilize It’s “About this image” function to determine if an image was made or changed using AI techniques if it has Google C2PA information. “About this image” is available in Google photos, Lens, and Circle to Search and helps provide users context for the photos they see online.
Ads: Google C2PA information is beginning to be integrated into Google ad systems. Their intention is to gradually increase this and utilize Google C2PA signals to guide the enforcement of important rules.
Later in the year, they’ll have more details on It’s investigation into how to notify YouTube users with C2PA information when material is recorded using a camera.
In order to enable platforms to verify the material’s provenance, Google will make sure that their implementations evaluate content against the soon-to-be-released Google C2PA Trust list. For instance, the trust list assists in verifying the accuracy of data if it indicates that a certain camera type was used to capture the picture.
These are just a few of the applications for content provenance technology that nous are considering at this moment. It want to add it to many more products in the future.
Determining and indicating the origin of material is still a difficult task that involves many factors depending on the item or service. Even if people are aware that there isn’t a single, universal solution for all online material, collaboration across industry players is essential to the development of long-lasting, cross-platform solutions. For this reason, it’s also urging more hardware and service providers to think about implementing the Google C2PA‘s Content Credentials.
It efforts with the Google C2PA are a direct extension of their larger strategy for openness and ethical AI research. For instance, it’s still adding Google DeepMind‘s SynthID embedded watermarking to more next-generation AI tools for content creation and a wider range of media types, such as text, audio, visual, and video. In addition, Google have established a coalition and the Secure AI Framework (SAIF) and joined a number of other organizations and coalitions devoted to AI safety and research. That are also making progress on the voluntary pledges they made at the White House last year.
Six up-and-coming artists from various backgrounds were asked to create new themes for the Chrome browser.Image Credit To Google
September marks the beginning of a season of change: a new school year, a new you, and matching Chrome themes.
Google started the Chrome-sponsored Artist Series a few years ago to honor the talent of artists worldwide and provide their creations as unique Chrome themes. They commissioned six brilliant up-and-coming artists from various backgrounds to present their work in Chrome for the newest collection, which is available beginning today: Melcher Oosterman, DIRTYPOTE, Kaitlin Brito, Kanioko, Kate Dehler, and Martha Olivia.Image Credit To Google
Check out the Rising Artists Series by visiting the Google Chrome Web Store. Select a theme that inspires you, click “Add to Chrome,” and take in the eye-catching hues and upbeat patterns. To see and use themes from this collection, you may alternatively create a new Chrome tab and click the “Customize Chrome” icon in the bottom right corner.
Read more on Govindhtech.com
DeepMind introduces a groundbreaking method to generate synchronized audio for silent videos, using deep learning.
This technology, named “AudioLM,” can create realistic and coherent sounds matching the visual activities in videos without using any textual input or pre-existing audio templates.
This advancement not only enhances the accessibility of videos for hearing-impaired viewers but also opens new avenues for content creation in the film and video industry.
.
.
.
For more AI related updates, follow @trillionstech.ai
🤖🔒 Dans un monde où la science-fiction devient réalité, Google DeepMind s’inspire des lois de la robotique d’Isaac Asimov pour créer une “constitution du robot” qui assure la sécurité humaine.
Une “constitution du robot” de Google pour ne pas faire de mal aux humains
Voici quelques points clés de cette avancée majeure :
– 🧠 **AutoRT**: Un système innovant pour former les robots, assurant une…
