OpenAI, the company behind the popular chatbot ChatGPT, has just announced a major update to its DALL-E image generator. Users can now edit specific parts of an AI-generated image, similar to software like Photoshop, thus providing much more precise control over the final result. This new feature is available to subscribers of ChatGPT Plus on the web, iOS and Android.
A giant step towards democratizing AI image creation
Until now, AI image generators like DALL-E produced unique, entirely new images with each request, even if the instructions remained the same.
This made it difficult to obtain an accurate and consistent result. With this new editing tool, users can select a specific area of the image and ask the AI to edit it according to their instructions, while leaving the rest intact.
For example, you can ask DALL-E to:
- change the color of an element,
- to add or delete an object,
- or even modify the expression of a character.
The possibilities are almost endless. This advancement represents a giant step towards democratization of AI image creationmaking it accessible to everyone, even without advanced technical skills.
Limits that remain to be overcome
However, despite this impressive progress, AI still shows certain limitations in understanding and interpreting queries. In our testing, we found that DALL-E could sometimes completely delete an element instead of modifying itor failing to add a requested object in a complex scene.
These errors reveal that the AI does not actually “understand” what it is generating. It is based on probabilities and pixel correspondences, without having a real notion of geometry, physical space or the logic of objects. This is why we often observe inconsistencies in generated imageslike buildings with impossible shapes or characters who disappear into nothingness.
Towards ever more powerful artistic AI
Despite these challenges, OpenAI does not intend to stop there. The company claims to be actively working to improve these editing tools, in order to make them ever more precise and reliable. Ultimately, the objective is to enable anyone to create photorealistic images or digital artwork in just a few clicks, without requiring drawing or photo editing skills.
This update to DALL-E marks a crucial step in the evolution of image-generating AI. While the technical challenges remain numerous, there is no doubt that in the months and years to come we will witness the emergence of ever more powerful tools, redefining our relationship with visual creation. An exciting revolution, both rich in opportunities and ethical questions about the place of AI in art.
My name is Ethan, I’m 30 years old, and I’m the founder of this magazine. After studying journalism, I quickly decided to go freelance to be able to write as I wanted and talk about my true passions. I aim to create an information hub where French and international news is treated seriously and with insight. My goal is to enlighten my readers’ daily lives and encourage a deeper understanding of current issues.
2024-04-06 11:30:00