Limitations of Dall-E 3 (Part 2)

I've been exploring the capabilities of Dall-E 3, an advanced AI image generator from OpenAI, in my recent blog posts. In the previous post, I discussed some of its limitations such as misunderstanding negative prompts and failing to remove certain objects like Las Vegas lights or the Hollywood sign in Los Angeles. In this part 2, I delve deeper into these issues and discover that Dall-E is overly paranoid about copyright, has trouble understanding cause and effect, and struggles with rendering specific locations accurately while hallucinating multiple versions of landmarks in the background. Despite these limitations, its images are still impressive and could be a valuable tool for generating building mockups once OpenAI addresses accuracy issues.

https://investomation.com/blog/limitations-of-dall-e-3-part-2