Other Technology | May 16, 2024
Editing has arrived for artificial intelligence image generation, allowing for the customization of prompt results.
After adding watermarks to DALL-E 3 in February, OpenAI has announced another surprising innovation to artificial intelligence image generation: an editor to modify the generated prompt according to user preferences. This means you can say goodbye to repeatedly entering different prompts to achieve the desired result.
AI image generation is based on entering a series of words known as prompts. If the final result is not what you’re looking for, you need to try entering other words to refine the composition, technique, or style of the design you’re aiming for.
But now everything changes with OpenAI’s announcement that includes an image editor to modify any area. While this isn’t entirely new—as Photoshop allows marking an area for Adobe’s AI to describe what to generate within it—the integration into an AI-driven image generation tool like DALL-E opens new possibilities for customization and creativity.
The key point here is that ChatGPT can assist in defining images created by DALL-E 3 with an image editor. After DALL-E 3 generates the image, you can use a tool to modify the desired parts. Immediately after, ChatGPT is used to ‘tell’ what you want to generate in that area.
In the example provided by OpenAI, the generation of a dog is shown, and with the use of the tool, two areas are marked where bows will be placed. ChatGPT is then used to instruct the addition of bows to the marked area. After a few seconds, a new image with the bows is generated by DALL-E 3.
Unlike Photoshop with its AI functions, OpenAI’s image editing or generative fill does not offer the variants that Adobe’s tool does. That is, once the bows are added, you could use up to three variants to find one that fits better.
However, this new functionality of ChatGPT with DALL-E 3 is currently only available through the ChatGPT Plus subscription, and it is expected that more options will be offered in the future, such as the ability to change a person’s facial expression or erase parts of a generated image.
