On September 2023, OpenAI announced new text-to-image model, Dall·E 3. I first got the oppurtunity to try Dall e 3 when my company give access to ChatGPT Pro, which include the Dall E 3 model. I am pretty excited because the image generated is far better than Dall·E 2 that I've explored with some times ago.
So today, I create a simple python program and incorporated the OpenAI's Image Generation API to redo the prompt from previous Dall·E 2 experiment. Note that the Dall·E 2 images shows below are generated on November 2022.
Below is the image comparison of the Dall·E 2 and Dall·E 3 side-by side. Sorry for the inconsistent image sizes.
1. A realistic image of a boss cat doing work in office
|Totally unsatisfying, the picture is not realistic at all.
|The picture is much more realistic and nicely composed. Also, cute.
2. A digital art of a lighthouse and with andromeda galaxy at the background
|Good, just a bit cartoonish
|Satisfied. Nicely composed. But it looks like earth has no atmosphere at all (The galaxy should looks faded, but I assume maybe it uses long focal length setting)
3. Realistic image of human civilization on Mars
|Ok good enough
|Better, have more life
4. A depiction of life in year 3040
|Looks stressful, apocalypse like scenario (maybe because this image is generated just after the spike of Covid-19 era)
|Future-modern life like depicted in most of my primary school textbooks
5. Profile picture with a man and cat in tech environment
|Good enough, but why the cat is sad though
|Not composed nicely. The cat looks more realistic than the human.
Malaysia Programmer (Generated via ChatGPT using Dall·E model)
Malaysia Prayer time logo for mobile app
Overall, I think Dall E·3 now are able to generate results that are richer in context and eye-pleasing. For those who want to see python script used to generate the image, please refer to GitHub repository openai-dalle2 & openai-dall-e3.