In the world of AI, advancements that break barriers we deemed impossible are happening at a very fast pace. OpenAI‘s DALL-E 3 model may just seem like yet another new model in the vast sea of AI tools today, yet it brings real innovation.
Ideas to images
DALL-E 3 comes with lots of changes. Many of them are behind-the-scenes improvements to how the model works, like new limits on violent or harmful content. Some improvements are those we can witness for ourselves. For one, DALL-E 2 and similar models produced images with incomprehensive text if asked to include it. The new model has been shown to produce images with readable text. While it has some typos, it’s still a big improvement from its predecessor.
Compared to midjourney or stable diffusion, Dalle-3 text support is truly a huge advantage for technology adoption and real-world usage of image generation… #dalle3 #stablediffusion #midjourney pic.twitter.com/3jstAFQSmc
— redd (@reddu4) September 30, 2023
However, a more debatable change is that DALL-E 3 will no longer be trained on data from live artists. There will also be restrictions around using public figures as a subject. Generative AI technology is still a grey area which is why many authors have been suing OpenAI over their works being used to train the models. As such, OpenAI has created an opt-out form for creators so they can personally make sure their data is excluded.
Prompts no more
DALL-E 3 doesn’t stop at model changes. A job description no one would have thought of a decade ago is a prompt engineer. Yet in the past year, this became a reality with the surge in popularity of generative AI tools. Specifically for image generation tools like Stable Diffusion and DALL-E itself, the prompt needs to be a descriptive and precise sequence of keywords.
DALL-E 3 can design website mockups in seconds.
This will change everything for designers — infinite ideas and inspiration. pic.twitter.com/wDb2wuMCk4
— Matt Shumer (@mattshumer_) October 4, 2023
Since normal language won’t produce exactly what you want, you would need to learn how to engineer this prompt. Alternatively, as the case with many companies that started using these tools professionally, hire someone specialized in optimizing the prompts. Starting with DALL-E 3 and with the integration of ChatGPT, this won’t be needed. Anyone will be able to just describe what they want however they want, and the AI will understand.
YouTube: Introducing DALL·E 3
Photo credit: The feature image is symbolic and has been done by Christopher Isak with Midjourney for TechAcute.
Source: Jennifer Zhan (Vulture) / Stable Diffusion Art
