Exploring DALL-E 3’s New Image Editing Tool: Lessons Learned & Tips for Success

Earlier this week, ChatGPT introduced DALL-E 3’s new editing tool. Today was the first day I had access to it, so I was eager to try it! I started by generating a picture of a man from the 1870s dressed in period-appropriate clothing and setting. Curious to test the tool’s capabilities, I attempted to remove a candle from the image’s foreground.

Accessing the Edit Tool

To access the edit tool, you first click on the image and then the paintbrush icon next to the save icon. Your cursor becomes a circle; you can adjust its size with a slider tool in the upper left.

Removing Objects: A Trial-and-Error Process

I roughly highlighted the candle—you don’t have to be exact—and it outlined the candle with a dotted line. See Image 2.

I then prompted: “remove candle.” DALL-E generated a new image, removing about 3/4 of the candle. See Image 3.

However, after several messages back and forth, as it claimed it had already removed the candle. Reflecting on this, I realized it had, in fact, removed the candle. What was remaining was part of the candlestick! So, I started over and prompted, “Remove the candle and the candlestick” after highlighting again. This time, the result was impressive! See image 4.

Learning to Communicate with AI

This experience underscored the importance of clear communication with AI, a topic I’ve explored in two previous posts. In “The Art of Prompting: How to Effectively Communicate with AI,” I covered five essential prompt types that can help you get the most out of your AI interactions. And in “Simple Prompts vs. Prompt Engineering: A Genealogist’s Experiment with AI,” I demonstrated that simple, plain English prompts (which I now know are called “vanilla prompts”) often work just as well as more complex, engineered prompts.

In this case, when I initially prompted “remove candle,” the AI only partially removed the object. However, when I rephrased my prompt to “remove the candle and the candlestick,” the tool performed excellently in object removal, demonstrating the power of precise instructions. This experience reinforces the ideas explored in my previous posts—that effective communication with AI is a skill that can be learned and refined over time, and that sometimes, the simplest approach is the most effective.

Comparing with Canva

I have used Canva for months, which excels at removing objects, usually fading them gradually over multiple attempts. However, I haven’t had much success with it in replacing objects or changing the color of an item. In an experiment with DALL-E’s tool to change the feather’s color to blue in the same image as the candle and candleholder, the results were underwhelming. The feather did turn blue, but it was tiny and looked odd. A subsequent attempt to enlarge the feather resulted in a large, fuzzy image that didn’t fit well with the rest of the picture. See image 5 for the tiny blue feather.

Concluding Thoughts

While DALL-E’s editing tool has shown promise in object removal, its performance in other tasks like color changing or resizing is less convincing. I encorage others to try it, especially for object removal, and share their experiences. With the rapid development of AI-driven image editing tools, genealogists and history enthusiasts have an exciting opportunity to enhance their visual storytelling by experimenting with these tools and mastering their features.

Have you experimented with DALL-E’s new image editing tool, or are you using Canva or another tool for your image editing? Please share your experiences and tips in the comments below. Let’s learn from each other and explore how AI can enhance our visual storytelling!

P.S. If you’re interested in learning more about how to effectively use AI to help you with both genealogy and everyday tasks, please check out my 8-week course, AI Explorations in Genealogy & Beyond, which starts April 18th. The class is almost full, and I’d love to have you join us!