Apple’s new AI can modify images based on natural language sentences


MGIE, the AI ​​model introduced by Apple, promises to transform image editing by interpreting text instructions, marking a major breakthrough in AI-assisted creative for the enterprise.

ia photos apple
Source: arxiv

The evolution of AI in the field of image creation has taken significant steps forward, notably with Microsoft’s Bing Image Creator, providing users with the ability to generate images from text descriptions. This innovation, along with subsequent improvements in speed and efficiency, demonstrates the company’s rapid advancement in the use ofartificial intelligence for some creative apps.

Microsoft recently overtook Apple in terms of valuation, mainly due to its commitment and significant advances in the field of artificial intelligence, as evidenced by success of ChatGPT-4 and Copilot Pro. In this context, Apple MGIE appears as a strategic response aimed at maintaining its competitiveness in the field of AI.

Apple introduces new AI image editing model

Result of a collaboration between Apple and researchers at the University of California at Santa Barbara., MGIE (MLLM-Guided Image Editing) stands out for its ability to understand and execute text commands for precise manipulations at the pixel level. This ability to transform textual instructions simple or ambiguous in clear and precise directives opens up new possibilities forintuitive image editing. Presented at the ICLR 2024 conference, this model demonstrates remarkable efficiency. It offers accurate results that have been approved by users, while working quickly and without wasting resources.

MGIE stands out for its ability to achieve wide range of modifications images, ranging from simple color adjustments to complex object transformations. The template also optimizes photos globally and allows for targeted retouching on specific areas. It excels at cropping, resizing, rotating images, as well as adjusting brightness, contrast, and color balance, all from simple text prompts. Available as open source, this program is easy to access via GitHub, with additional resources like a demo notebook and web demo at Hugging Face Spacesmaking it convenient to use for various editing applications.

Source: arxiv



Source link -101