DALL-E 3 brings several enhancements. It is now more proficient in responding to comprehensive and detailed prompts, resulting in the creation of “striking” images. The tool also offers support for both landscape and portrait aspect ratios. Behind the scenes, the model pays closer attention to user-generated captions, improving its ability to fulfill user instructions effectively.
OpenAI is also working on a digital provenance tool to verify the authenticity of DALL-E generated images. Initial tests have demonstrated a remarkable 99% accuracy in identifying unedited DALL-E images, which remains a respectable 95% even when minor edits like cropping or compression are applied.
