Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present the hot topics in the AI field, focusing on developers, helping you understand technical trends and innovative AI product applications.

Fresh AI products click to learn: https://top.aibase.com/

1. Midjourney launches model personalization feature

Midjourney recently introduced an innovative model personalization feature that allows users to fine-tune the MJ model according to their aesthetic preferences, making the generated images more aligned with personal tastes. The personalized model achieves this by analyzing the user's liked images, learning the user's unique aesthetics, and catering to the user's preferences. Users need to rate or like at least 200 images. After enabling the personalization feature, they can add a specific code at the end of the prompt to share the personalized effect.

image.png

【AiBase Summary:】

🎨 Model personalization feature allows users to fine-tune the MJ model according to their aesthetic preferences, making the generated images more aligned with personal tastes.

🔍 Personalized model achieves this by analyzing the user's liked images, learning the user's unique aesthetics, and catering to the user's preferences.

💡 Users need to rate or like at least 200 images. After enabling the personalization feature, they can add a specific code at the end of the prompt to share the personalized effect.

Details link: https://www.midjourney.com/rank

2. ComfyUI completes adaptation in advance for the upcoming open-source SD3 Medium model

The SD3 Medium model is about to be open-sourced. To prepare for this significant moment, ComfyUI has completed the adaptation work in advance, ready to welcome the new model. At that time, everyone can experience more realistic textures, better composition, superior performance, and better fine-tuning capabilities when generating images.

QQ Screenshot 20240611174410.jpg

【AiBase Summary:】

📈 The SD3 Medium model is about to be open-sourced, and ComfyUI has completed the adaptation work in advance.

🖼️ The SD3 Medium model has made significant progress in image generation, capable of generating images with rich details and high realism.

💡 The SD3 Medium model has strong capabilities in generating high-quality, detailed images.

Details: https://github.com/comfyanonymous/ComfyUI/commit/8c4a9befa7261b6fc78407ace90a57d21bfe631e

3. WeChat keyboard beta adds AI assistant feature, get AI-generated answers by pressing this key

The WeChat keyboard beta has added an AI assistant feature, allowing users to get AI-generated answers by pressing the "=" key. It also supports emoji and emoticon recommendations, as well as enhanced support for time and date input formats.

image.png

【AiBase Summary:】

🤖 WeChat keyboard adds AI assistant feature, get AI-generated answers by pressing the "=" key.

🤖 Updated to support emoji and emoticon recommendations, enhanced support for time and date input formats.

🤖 Currently in beta version for Windows users, iOS, Android, and Mac platforms may be released in the future.

4. ByteDance launches AI virtual dating chat product Banana – can generate photos and chat like a real person

Recently, an AI virtual dating chat product named "Banana" (Chatwiz in English) was launched, featuring the ability to generate photos and chat very close to a real person. After verification, the product is owned by Beijing Zhendi Technology Co., Ltd. (Tomato Novel), and its actual controller is ByteDance.

QQ Screenshot 20240611160830.jpg

【AiBase Summary:】

⭐ Banana is an AI virtual dating chat product that can generate photos and chat close to a real person.

⭐ ByteDance is actively deploying AI large models, launching multiple AI products and services.

⭐ "Banana" demonstrates ByteDance's continuous exploration and innovation in the AI application field.

5. iFlytek to release iFlytek Spark V4.0 on June 27, showcasing the latest intelligent voice technology

iFlytek will release the iFlytek Spark V4.0 on June 27, showcasing the latest end-to-end intelligent voice technology achievements, including one-sentence replication, high-noise scene speech recognition, and multi-dialect multilingual seamless switching. Liu Qingfeng revealed that iFlytek is at the international leading level in full-duplex technology and hyper-realistic synthesis technology. In the future, iFlytek will focus on the development of far-field high-noise multi-person speaking scenes, high-expressiveness personalized scenes, and other fields.

【AiBase Summary:】

🚀 iFlytek Spark V4.0 will showcase the latest end-to-end intelligent voice technology achievements, including one-sentence replication, high-noise scene speech recognition, and multi-dialect multilingual seamless switching.

💡 iFlytek is at the international leading level in full-duplex technology and hyper-realistic synthesis technology.

🔮 In the future, iFlytek will focus on the development of far-field high-noise multi-person speaking scenes, high-expressiveness personalized scenes, and other fields.

6. Apple's stock hits record high after announcing new AI features

Apple's stock closed up more than 7% on Tuesday, hitting a record high. This rebound brought hope for Apple's performance this year, showing the market's positive attitude towards Apple's new artificial intelligence features.

image.png

【AiBase Summary:】

📈 Apple's stock rose more than 7% on Tuesday, hitting a record high, with a market value expected to reach $3.18 trillion, second only to Microsoft.

📱 New AI features have increased the attractiveness of Apple devices, including improved Siri virtual assistant and multiple AI features.

💡 After the developer event, analysts raised their target price for Apple stock, expecting new features to stimulate purchases of the new iPhone series in the fall.

7. Follow-Your-Emoji: Generate expressive animations by capturing facial expressions

Follow-Your-Emoji is a breakthrough technology that generates new facial animations by extracting facial features from videos. This technology accurately captures facial features and pupil points, excluding facial contours, to achieve more natural and vivid animation effects. It has a wide range of applications, benefiting the entertainment, education, and business sectors.

image.png

【AiBase Summary:】

👤 Users provide photos, and the technology generates video animations, capturing subtle facial expression changes.

🔒 Identity is maintained, with the reference avatar's identity features preserved and not lost.

😊 Rich expressions, generating various expressions, including pupil movement, making the animation more lively and realistic.

Details link: https://top.aibase.com/tool/follow-your-emoji

8. Online AI image editor Freepik Designer

Freepik Designer is an innovative online AI image editor that provides users with simple and easy-to-use design tools, allowing for quick mastery without professional design skills. Its integrated AI tools make the design process more efficient, while also offering a rich template library to meet different design needs.

image.png

【AiBase Summary:】

🎨 Simple and easy-to-use design tools, allowing for quick mastery without professional design skills

🖼️ Offers a rich template library to meet different design needs

💡 Integrated AI tools to enhance design efficiency and quality

Details link: https://top.aibase.com/tool/freepik-designer

9. Elon Musk withdraws lawsuit against OpenAI

Elon Musk has withdrawn his lawsuit against OpenAI, accusing it of breach of contract. Musk believes that OpenAI has abandoned its non-profit mission in favor of commercial interests. OpenAI denies the allegations, calling them "incoherent" and "absurd".

【AiBase Summary:】

🔍 Elon Musk withdraws his lawsuit against OpenAI.

💡 Musk accuses OpenAI of abandoning its non-profit mission in favor of commercial interests.

🔒 OpenAI denies the allegations, calling them "incoherent" and "absurd".

10. Yandex releases open-source tool YaFSDP, breaking through the efficiency bottleneck of LLM training

Yandex's open-source YaFSDP tool has brought a breakthrough in LLM training optimization methods to the global AI community, significantly improving training speed and saving a large amount of GPU resources, making autonomous LLM training more feasible. Yandex is committed to continuously contributing to the development of the global AI community, and the open-source YaFSDP is a reflection of this commitment.

【AiBase Summary:】

✨ YaFSDP is an efficient large language model training optimization method released by Yandex, capable of improving LLM training speed by 26%.

💡 YaFSDP focuses on optimizing GPU communication efficiency and memory usage, performing excellently when training parameters scale from 30 billion to 70 billion.

🌟 Training a 70 billion parameter model with YaFSDP can save about 150 GPU resources, with cost savings ranging from $500,000 to $1.5 million.

11. Speed increased by 410 times! TiTok can reconstruct and generate images with just 32 tokens

Recent advancements in generative models have highlighted the crucial role of image tokenization in the efficient synthesis of high-resolution images. TiTok is a Transformer-based one-dimensional tokenization framework that tokenizes images into one-dimensional latent sequences, greatly improving generation efficiency and quality. It excels in processing high-resolution images, significantly increasing generation speed while maintaining high-quality sample output.

image.png

【AiBase Summary:】

⚙️ Image tokenization reduces computational needs, enhancing generation efficiency and effectiveness.

🔍 TiTok tokenizes images into one-dimensional latent sequences, representing 256×256 images with as few as 32 discrete tokens.

💡 TiTok performs excellently on ImageNet benchmarks, increasing generation speed by 410 times while maintaining high-quality sample output.

12. MIT develops new algorithm DenseAV: Learning the meaning of language by watching videos

In the new DenseAV algorithm developed by MIT, researchers use machine understanding to learn the meaning of language by watching videos of animal communication. The algorithm can learn the meaning of words and the location of sounds in an unsupervised manner, achieving a natural distinction in cross-modal connections. The team hopes to apply this to understanding new languages and discovering patterns of association between different signals.

image.png

【AiBase Summary:】

🧠 DenseAV is a dual-encoder grounding architecture that learns features with high resolution, semantic meaning, and audiovisual alignment.

🔍 Unsupervised learning discovers associations between word meanings and sound locations, automatically distinguishing between language and sound.

🌐 Outperforms previous models like ImageBind in cross-modal retrieval, applicable to learning from a large number of videos and understanding new languages.

Details link: https://top.aibase.com/tool/denseav

13. Making AI more ethical: Source.Plus provides high-quality AI training data

Spawning is committed to providing artists with more control over the online use of their work. The Source.Plus project has released a dataset containing nearly 40 million public domain images and images licensed under Creative Commons CC0, providing high-quality data for AI model training. The platform provides artists and creators with more refined management of the usage rights of their works, injecting new vitality into the development and application of AI technology.

image.png

【AiBase Summary:】

🔍 Data search and organization: users can quickly search for various media data and organize and annotate it to meet training needs.

🌟 High-quality training data: data that has been screened and reviewed to ensure safety and quality, with legal permission to use.

💡 Wide range of applications: suitable for various AI model training, improving accuracy and robustness.

Details link: https://top.aibase.com/tool/source-plusSource.Plus

14. Mistral AI raises $640 million in Series B funding

Mistral AI recently announced the completion of a $640 million Series B financing round, with a valuation of nearly $6 billion. This round of financing was led by General Catalyst, with participation from several well-known investment institutions and companies, accelerating Mistral's development in the field of artificial intelligence and its international commercialization process.

【AiBase Summary:】