DALL-E 3: a step towards content policy moderation | by Andrea Valenzuela | Nov, 2023


Understanding the new AI image generators and their risks

Animated gif of a AI robot painting art.
Self-made gif.

OpenAI launched last month, on Friday 20th of October, the new version of the image generator model DALL-E 3. This updated version is now accessible to all ChatGPT Plus & Enterprise users, featuring an integration directly within ChatGPT that allows users to create original images through conversation.

However, DALL-E 3 isn’t entirely new to the public; it has been the driving force behind the Bing Image Generator, where the user interface largely mirrors that of DALL-E 2.

Despite the underlying models being technically identical, I’ve been playing around with both of them, and they really perform differently! In this article, I aim to share my first findings and impressions.

The user experience of DALL-E 3 differs from its predecessor, DALL-E 2, due to its conversational interface. In this integration, users interact with ChatGPT rather than crafting prompts for the image generator themselves. ChatGPT takes the essence of the user’s conversation and formulates a prompt on their behalf.

This innovative approach to prompting addresses certain user interaction challenges encountered with DALL-E 2. As in ChatGPT, the quality of the model output also depends on the quality of the user prompt. With ChatGPT as an interface between the user and the model, prompts are much more curated allowing the model to generate better images.


Source link

2023. All Rights Reserved.