The Intelligent AI Assistant Doubao, developed by ByteDance, has recently upgraded its text-to-image capabilities. The new feature allows users to generate images containing specified text with a single click. The relevant person in charge of the Doubao large model team pointed out that the text-to-image model of Doubao integrates LLM (Large Language Model) and DIT (Digital Image Technology) framework, optimizing its learning ability for Chinese data, and enhancing the generation capability of Chinese characters, significantly improving the generation results.

The rapid development of generative AI has raised the challenge of comprehensively evaluating its performance. A plethora of models have emerged, each showcasing increasingly impressive results. However, the pressing question remains: how do we assess the effectiveness of these text-to-image models? Traditional evaluation methods often rely on subjective visual assessments or simple metrics like CLIPScore, which frequently fail to capture the intricate details found in complex text prompts, such as relationships between objects and logical reasoning. This leads to inaccurate evaluation outcomes for many text-to-image models.

Recent research from Alibaba's Tongyi Laboratory indicates that existing text-to-image Diffusion Transformer models already possess the ability to generate multiple images with specific relational contexts. With just a little 'guidance', they can achieve a synthesis of understanding and produce high-quality multi-image collections. Traditional Diffusion models are more like 'rote learning' students, requiring massive datasets to generate high-quality images. However, with the support of IC-LoRA, they become 'high-achieving students', capable of learning through analogy.

Last night, Stability AI unveiled its most powerful model - Stable Diffusion 3.5, which is not just a single model, but a package containing three versions, designed to meet the diverse needs of researchers, business enthusiasts, startups, and enterprises. The three versions are Stable Diffusion 3.5 Large, Stable Diffusion 3.5 Large Turbo, and one that will be available on October 29.

The Zhipu technology team announced an exciting piece of news today: their latest text-to-image model CogView3 and its upgraded version CogView3-Plus-3B have officially been open-sourced and successfully launched on the 'Zhipu Qingyan' App. The debut of these two models marks a new phase in AI-assisted artistic creation. CogView3 is a cascade diffusion-based text-to-image model, and its generation process is exquisite. The model first generates a low-resolution image of 512x512 pixels, and then...

Recently, Zhichu AI has open-sourced its latest creation—CogView3 and its upgraded version CogView-3Plus-3B, injecting new vitality into the field of text-to-image generation. The debut of CogView3 is undoubtedly a significant milestone. As the first model to achieve relay diffusion in the text-to-image generation domain, it employs a unique cascade diffusion method. This innovative approach first generates low-resolution images and then completes the final output through relay-based super-resolution technology. This not only greatly enhances...

Baidu Input Method announces a comprehensive upgrade of its all-scenario AI writing assistant 'Super Writing', integrating a variety of functions such as AI writing, AI chatting, and AI intelligent Q&A, enhancing its comprehensive strength as an AI input method product. In addition, Baidu Input Method has also launched new multimodal creation features like 'Text-to-Image', further improving users' content creation experience in different scenarios.

Stability AI collaborates with Amazon Web Services (AWS) to introduce three new text-to-image generation models to the Amazon Bedrock service, aiming to enhance business positioning and market competitiveness. These three models—Stable Image Ultra, Stable Diffusion3Large, and Stable Image Core—cater to different needs, offering everything from high-quality image generation to fast bulk image production.

On August 21, 2024, the Ideogram team proudly announced the launch of its latest version of the text-to-image model—Ideogram 2.0. This significant update not only provides users with free access but also introduces a series of innovative features and tools, marking a major leap in text-to-image technology.

Google's text-to-image model Imagen3's 'ImageFX' feature is now officially available to all users, offering an enhanced prompt interaction experience. Users only need to input descriptive text, and the system can generate corresponding images while optimizing the prompts, providing various options to improve the generation results. Furthermore, Imagen3 also includes a localized modification feature, allowing users to use brush tools to edit specific parts of the image and fill in related descriptive terms, with the system automatically generating the modified image, greatly improving the flexibility and convenience of image generation.

The Google research team has released a groundbreaking technology called Alchemist, allowing users to precisely edit material properties of objects in images, such as color, gloss, and transparency, without the need for professional software or skills. The core of the technology is a 'Text-to-Image' generation model that learns to change specified material properties while keeping the shape and lighting unchanged by fine-tuning and creating synthetic datasets (including 3D models, various materials, camera angles, and lighting conditions). Experiments show that this technology improves the photorealism of images.

At the Samsung Mobile China New Product Launch on July 17th, Meitu Inc. and Samsung China announced an enhanced collaboration. Meitu's MiracleVision large model will provide "text-to-image wallpaper" and "doodle-to-image" functions for users of Samsung's Galaxy Z Fold6 and Z Flip6 series phones. These features allow users to generate creative wallpapers that integrate artistic aesthetics by selecting themes, styles, and other keywords, or by using the S Pen to scribble on the panel and transform

According to reports from Titanium Media, the ByteDance team plans to unveil significant advancements in their latest artificial intelligence model technology on July 19th (this Friday) for the first time on a large scale. The content to be disclosed includes new AI models such as text-to-image and Sora-like video generation, with particular emphasis on innovative technologies in long videos and high-dynamic directions, which are expected to directly compete with OpenAI's Sora text-to-video mode

Developed jointly by researchers from the University of Hong Kong and Bytedance, LlamaGen is an image generation method based on the autoregressive model Llama, which has shown the potential to surpass traditional diffusion models in the field of image generation. The open-source release of LlamaGen quickly garnered nearly 900 stars on GitHub, acknowledging this achievement. This result not only proves the competitiveness of autoregressive models in image generation but also brings new vitality

Recently, the AI research organization Artificial Analysis launched an initiative called 'Artificial Analysis Text to Image Leaderboard & Arena' to comprehensively evaluate the performance of these models.

UK-based AI startup Stability AI has appointed Prem Akkaraju as its new CEO. Akkaraju previously served as the CEO of visual effects company Weta Digital and is also one of Stability AI's investors. The investor lineup for Stability AI is impressive, including former Facebook president Sean Parker among others, with total investments exceeding 1.

Douyin's Flow department has launched an AI camera product called 'Xinghui', offering a brand-new visual experience. 'Xinghui' supports AI avatars, allowing users to customize a variety of avatar effects. Users can generate their own AI avatars by uploading or taking photos, navigating through the virtual world. 'Xinghui' also provides features such as outfit coordination, career simulations, and hairstyle changes to make AI avatars more personalized. In addition to AI avatars, users can also use 'Xinghui' for AI retouching and text-to-image creation.

The Stable Diffusion 3 model introduces the DiT (Diffusion Transformer) architecture, which significantly improves quality. Stable Diffusion 3 outperforms other text-to-image generation systems. The parameter count of Stable Diffusion 3 ranges from 800M to 8B. The SD3 architecture is developed in collaboration with core members of Sora and an assistant professor from New York University.

Stable Diffusion 3 is the most powerful text-to-image model. It adopts the MMDiT architecture, demonstrating performance that surpasses existing text-to-image generation systems. Stable Diffusion 3 excels in visual aesthetics, text adherence, and layout compared to other advanced models. The MMDiT architecture combines DiT and rectangular flow forms, processing image and language representations through independent weight sets. Stable Diffusion 3 offers flexibility.