ChatGPT creates hyper-realistic images, plots graphs for users

During a livestreamed event on March 25, OpenAI demonstrated how users could use ChatGPT to edit images simply by talking to the chatbot. For example, they could ask ChatGPT to create a photo of a snail in a city and then add details like a hat or change the background.

An image generated by ChatGPT. Photo: OpenAI

OpenAI says ChatGPT can generate better images when given clear input, and can help draw graphs and design logos for work. Users can ask the chatbot to create menus, maps, or perform complex instructions on image layout.

Aiming to become a general-purpose application, OpenAI is integrating ChatGPT with features like search, voice assistant, and video creation. The latest improvements help the product compete with other chatbots, including Elon Musk's Grok, and meet the needs of both individuals and businesses.

Still, ChatGPT’s image generation is not perfect, for example, it can make spelling mistakes in images. According to OpenAI, these errors often occur if the user’s prompts are not specific enough. The AI ​​also has trouble generating small characters or letters outside the Latin alphabet.

According to CEO Sam Altman, ChatGPT will take about a minute to generate an image due to the higher level of detail. GPT-4o will replace DALL·E as the default image generation model on ChatGPT, and is available to all users, including ChatGPT Free, Plus, Team, and Pro.

OpenAI says the advances in GPT-4o — which it released last year — were made possible by a team of “human coaches” who labeled the model’s training data.

However, the use of AI to generate images remains controversial, with some artists concerned that these tools “copycat” their work and threaten their livelihoods. OpenAI says GPT-4o was trained on public data and proprietary sources from partners like Shutterstock.

(According to WSJ, Bloomberg)

Source: https://vietnamnet.vn/chatgpt-tao-anh-sieu-thuc-ve-do-thi-cho-nguoi-dung-2384665.html