Welcome to the AI Daily column! This is your daily guide to exploring the world of artificial intelligence. Every day, we bring you the hottest topics in the AI field, focusing on developers, helping you stay on top of tech trends and understand innovative AI product applications.

Fresh AI Products Click to Learn More: https://top.aibase.com/

1. Luma Official Releases Video Introducing Dream Machine Model Features

After watching the video introduction of the Dream Machine model released by Luma, I feel that this AI video generation tool is very powerful. It not only provides high-quality video output but also can quickly understand user prompts to generate video content that aligns with aesthetic styles. This is very helpful for the creative iteration process, making video generation more efficient.

AiBase Highlights:

🌟 High-quality video generation with a resolution of up to 1024 pixels.

🎨 Capable of understanding prompts to generate videos that align with aesthetic styles.

⚡ Fast inference speed, beneficial for rapid creative iteration.

Details Link: https://top.aibase.com/tool/dream-machine

2. New Lip Sync Video Project Hallo Released, Allows Precise Control Over Facial Expressions and Mouth Shapes

The new lip-sync video project Hallo has been released, which generates singing and speaking videos from a single image and audio input, achieving precise control over character expressions and postures, enhancing the alignment accuracy between voice input and generated animation. This technology can be used not only for virtual character animation generation but also for real characters, supporting multiple motion controls, cross-actor applications, and singing animation generation. It is technologically advanced, with realistic animations, and has broad application potential.

image.png

AiBase Highlights:

⭐️ Generates singing and speaking videos based on a single image and audio input.

⭐️ Supports virtual and real character animation generation, the project is open-source.

⭐️ Multiple motion controls, achieving precise expression and posture control, enhancing the diversity and realism of animations.

Project Address: https://top.aibase.com/tool/hallo

3. Peking University and Kuaishou Jointly Release Video Generation Framework VideoTetris, Surpassing Pika in Complex Video Generation

This article introduces the VideoTetris framework proposed by Peking University and the Kuaishou AI team to tackle the challenge of complex video generation, successfully surpassing commercial models Pika and Gen-2. The framework defines a combined video generation task, supports complex instructions and long video generation, and retains positional information and detailed features. The team employs a spatio-temporal combined diffusion method, optimizes training data preprocessing, and introduces a reference frame attention mechanism, generating more dynamic and natural videos.

image.png

AiBase Highlights:

⭐ VideoTetris framework successfully tackles the challenge of complex video generation, surpassing commercial models Pika and Gen-2.

⭐ Defines a combined video generation task, supports complex instructions and long video generation, retains positional information and detailed features.

⭐ Employs a spatio-temporal combined diffusion method, optimizes training data preprocessing, and introduces a reference frame attention mechanism, generating more dynamic and natural videos.

Details Link: https://top.aibase.com/tool/videotetris

4. Japanese AI Artist Revives Wife with Luma, Moves Netizens to Tears

This article tells the story of 65-year-old AI artist Matsuo Matsuo using technology to revive his wife Tori-chan, who passed away 11 years ago, touching countless people. Through AI technology, he re-orchestrated and recorded his wife's songs, created dynamic videos, and extracted and translated her letters, expressing his longing and love for his wife. This is a story of an ordinary person using technology to dream, showing the power and warmth of love in the AI era.

AiBase Highlights:

🌟 Matsuo Matsuo uses Luma's AI video Dream Machine to revive his wife Tori-chan, who passed away 11 years ago, moving many.

🎶 He re-orchestrated and recorded his wife's songs through AI technology and created dynamic videos, showing deep longing for his wife.

💖 Extracted and translated letters his wife wrote to him through AI tools, expressing profound love and eternal longing for his wife.

Product Entry: https://top.aibase.com/tool/luma-ai

Detailed Article: https://www.chinaz.com/ainews/9623.shtml

5. Apple's AI Plan Delayed, Developers to Test Only by Late Summer

According to Bloomberg, Apple's artificial intelligence (AI) plan will be a long and slow process. Apple's announced Apple Intelligence plan is expected to be available for developer testing only by late summer. This means it will not be among the first beta releases of Apple's new operating system updates and will only have a preview version released this fall.

AiBase Highlights:

🍏 Apple's artificial intelligence (AI) plan will be available for developer testing by late summer

📉 Apple's plan will not be among the first beta releases of new operating system updates

💬 Apple's Intelligence plan will bring changes to how consumers interact with devices and shop

6. KREA AI Launches Video Enhancement Feature, One-Click Boost to Video Quality

This article introduces the video enhancement feature "Enhancer" launched by KREA AI, which can improve the quality of images and videos and support higher resolutions and frame rates. Users can simply upload the target image/video that needs improvement, and KREA AI will process it online, ultimately generating high-quality videos.

AiBase Highlights:

⭐ Enhancer feature is open to everyone, can improve the quality of images and videos

⭐ Can be used in conjunction with AI video tools, after processing, you can directly view the comparison of enhanced effects on the page

⭐ Can generate videos with up to 2.5x pixels and frame rates as high as 120fps.

Product Entry: https://top.aibase.com/tool/krea-ai

7. Tsinghua and Peking University Collaborate to Release Long Video Understanding Benchmark Test: LVBench

This article introduces the long video understanding benchmark test project LVBench launched by Zhigu, Tsinghua University, and Peking University, aiming to address the challenges faced by multi-modal large language models in handling long videos. The project includes QA data for several hours across multiple categories, covering different types of video content, aiming to drive technological breakthroughs and innovation in the field of long videos. Many research institutions have already started working on the LVBench dataset, injecting new vitality into the fields of video understanding and multi-modal learning.

AiBase Highlights:

🔍 LVBench project is a long video understanding benchmark test project, including QA data for several hours across multiple categories.

💡 LVBench dataset covers various tasks such as video summarization, event detection, character recognition, and scene understanding.

🚀 The launch of the LVBench benchmark will drive breakthroughs and innovation in related technologies, injecting new momentum into the development of the long video field.

Details Link: https://github.com/THUDM/LVBench

8. Mesh Generation Model MeshAnything: Convert Any 3D to Artist-Created Meshes

Recently, 3D assets created by reconstruction and generation have reached the quality level of handcrafted assets, highlighting their potential in alternative fields. MeshAnything is an autoregressive model for generating artist-created 3D meshes, achieving high-quality mesh generation through VQ-VAE and a shape-conditioned decoder transformer. This method significantly improves storage, rendering, and simulation efficiency while maintaining comparable accuracy to previous methods.

image.png

AiBase Highlights:

⚙️ MeshAnything uses an autoregressive model to generate high-quality artist-created 3D meshes.

🔍 MeshAnything's meshes improve storage, rendering, and simulation efficiency while maintaining accuracy.

🌐 MeshAnything has a wide range of application scenarios across various fields, meeting different users' creative and needs.

Details Link: https://top.aibase.com/tool/meshanythingMeshAnything

9. Harvard Neuroscientists and Google DeepMind Create Artificial Brain in Virtual Mouse

This article introduces a groundbreaking study by Harvard University researchers in collaboration with Google DeepMind, using artificial intelligence technology to create an artificial "brain" for a virtual mouse. They successfully established a biologically realistic 3D mouse model and trained an artificial neural network brain using DeepMind's deep reinforcement learning algorithm, achieving simulation effects that surpass reality. This innovation is expected to bring revolutionary progress to the fields of neuroscience and artificial intelligence.

image.png

AiBase Highlights:

🧠 The virtual mouse has an artificial "brain" that can precisely control movement in complex environments.

🔬 The artificial neural network brain trained using DeepMind algorithms can generate various complex motion trajectories and forces.

🤖 The future application prospects are broad, potentially pioneering a new field of "virtual neuroscience" and bringing new strategies for the treatment of neurological diseases.

10. McDonald's Announces End of AI Drive-Thru Ordering Partnership with IBM

McDonald's announces the end of its AI drive-thru ordering partnership with IBM, and will remove the technology tested in over 100 restaurants by July 26, 2024. Although it is currently unclear why McDonald's is ending its partnership with IBM, the company stated that it is testing whether a voice-ordering chatbot can speed up service and expressed confidence in the test results. The restaurant industry is generally eager to introduce AI technology to improve efficiency.

AiBase Highlights:

🍔 McDonald's will end its AI drive-thru ordering partnership with IBM, removing the technology tested in over 100 restaurants.

🤖 McDonald's is testing a voice-ordering chatbot to speed up service.

🔮 The restaurant industry is generally eager to introduce AI technology to improve efficiency.

11. Study: People Struggle to Distinguish Between ChatGPT and Human After Five Minutes of Conversation

Large language models (LLMs) such as the GPT-4 model on the chat platform ChatGPT exhibit astonishing abilities, making it difficult to distinguish whether the generated text is written by a human. A study by the University of California, San Diego, found that people have a hard time distinguishing whether they are talking to a human or GPT-4, showing the extent to which machines can exhibit human intelligence.

image.png

AiBase Highlights:

🔍 GPT-4 model exhibits human-like conversation abilities that are difficult to distinguish in research.

💡 Study results show that in about 50% of interactions, people mistakenly think GPT-4 is human.