The author also explores the "leaked" DALL-E prompt, which reveals OpenAI's policies for image generation, including avoiding images of politicians or public figures, diversifying depictions of people, and modifying descriptions that include specific people. The prompt also outlines the default behavior of the model and provides instructions for handling various scenarios. The author concludes by noting that the prompt offers valuable insights into OpenAI's prompt engineering process.
Key takeaways:
- The author discusses their experience with DALL-E 3, an image generation model from OpenAI, which is accessed via ChatGPT and can generate different styles of images based on user prompts.
- DALL-E 3 has an interesting feature where it can maintain consistency between images given the same seed and a slightly altered prompt, which is not common in other image generation models.
- The author also explores the leaked DALL-E prompt, which reveals the inner workings of the DALL-E/ChatGPT combo and the policies followed to ensure the generation of diverse, inclusive, and non-offensive images.
- The article provides insights into prompt engineering within OpenAI, highlighting how the prompts have been developed and refined over time to counter harmful or embarrassing capabilities of the underlying DALL-E 3 model.