DALL-E has a new image-editing feature, allowing users to alter parts of an image instead of regenerating the entire thing. It might seem minor, but it’s a potential game-changer for content creators, allowing them to make targeted tweaks without having to start from scratch, saving them tons of time, and (hopefully) setting a new standard for what users expect from AI tools.
One of the most helpful AI tools for writers, marketers, and even some journalists are image generators. When your primary goal is to write articles, your accompanying imagery is often an afterthought, something to quickly grab from either an image vendor (e.g Getty, or Unsplash if you have no budget) or just screenshotted. For abstract concepts, though, it’s hard to find something that’s a good match for your topic, and most stock photos are, well, too stock.
That’s why services like Midjourney, DALL-E, and Stable Diffusion are so handy. Just prompt it with what you want, and you’ll often get a much better match for your article.
As anyone who has used these services knows, however, they’re not quite perfect. Putting aside engineered weirdness like what happened with Google Gemini, the output from image generators can often be off-target: the person you ordered up may look great, but they’re facing the wrong way. Or the landscape scene has all the right elements, but it’s the wrong season.
You can, of course, iterate on the output with more prompting, but that almost always regenerates the entire image, and you end up losing the parts of the image that worked. Whenever I’ve done this, it usually results in spending far too much time iterating again and again to get something like the first image until I just give up and start again (or settle for the initial imperfections).
Now with DALL-E, you can iterate on just part of the image, so when the AI gets things mostly right, you can preserve your overall picture while discarding the annoying bits. This potentially saves tons of time, but it also starts to blunt the common criticism of image- and video-generating AI tools — that they’re usefulness is limited by the grab-bag nature of their results.
When OpenAI announced Sora, for example, the video-generating tool was impressive for how easily it created realistic scenes, but there were plenty of strange results, too: What exactly can you do with a clip of a dog running around when it shows several other dogs “emerging” and disappearing? That’s a lot of work, for a human, in Premiere Pro or Final Cut just to clean it up. But what if you could just circle the multiplying dogs with your mouse right in Sora itself and tell the tool to get rid of them?
While OpenAI hasn’t announced any plans to introduce a similar editing ability in Sora, it seems like the next logical step, especially considering Adobe’s generative fill for Photoshop does something similar.
Looking at the DALL-E editor, however, a video equivalent may be a ways off. Out of the box, the current tool is very good at switching up details — for example, changing the tie in this picture of a robot newscaster from black-and-white to red — but incapable of adding entirely new elements. When I marked a section of the image and asked it to add another robot in either the background or foreground, it simply didn’t do the task, even though the response text claimed that it had.
Still, as a 1.0 on a feature, it’s not bad. Considering how often AI image generators get, say, the number of fingers or limbs on humans wrong, it will certainly save content creators tons of time. And it points the way for a new standard for smart features that allow for tweaking on AI output without entirely regenerating it — something surprisingly (and frustratingly) hard to get right with just prompting. It turns out when it comes to getting an AI to do exactly what you want, showing is better than telling.
Looking to understand how AI can help make you a more productive journalist, PR professional, marketer, or content creator? We highly recommend our thorough three-hour AI Bootstrapping for Marketers and Media class, happening April 16. Starting with foundational AI concepts, the class teaches the essentials of prompting, explores a suite of tools curated for creative work, and shows how to get started on custom solutions for your own workflows.
For those with busy schedules, we also offer a one-hour Beginning AI for Marketers, PR, and Journalists, happening April 15. It’s a crash course meant to rapidly bring novices up to speed on using generative tools. Even if you think you know the basics of ChatGPT and other chatbots already, this class will improve your use of it with a focus on advanced prompting techniques, underutilized features, and a set of go-to tools for speeding up work.
As a newsletter subscriber, you can take advantage of an early-bird discount on both classes with the code AISPRING, which gives you 50% off either class. The discount code will only work through April 10, so be sure to register before then. Here’s that link again to start your registration.
Leave a Reply