Midjourney Unveils Powerful AI Image Editing Tool, Raising Excitement and Ethical Concerns

Midjourney wants everyone to be able to edit images thanks to AI - Softonic

Midjourney, the well-known AI image generation platform, has announced the launch of an enhanced web tool that will allow users to edit any uploaded image from their page. According to David Holtz, CEO of Midjourney, this new feature will be available early this week and will enable advanced edits, such as retexturing objects in images to 'repaint' their colors and details just by giving instructions. The enhanced editing tool will be available only to 'a subset of the current community' initially, and will have stricter moderation, both human and through 'new more advanced AI moderators' to prevent potential abuses, as noted by Holtz in a post on his official Discord server. Image editing with artificial intelligence has become quite a controversial topic in recent months. Google, for example, has already launched quite powerful AI editing features that currently do not provide visual indications that the images have indeed been modified by AI. On the other hand, Meta continues to research how to label AI-edited images versus those generated from scratch. In this context, Midjourney stands out for its commitment to using Digital Source Type, from IPTC, a technical standard that incorporates metadata into images to indicate if they have been generated by AI. Even so, the company has not adopted the C2PA technology, which tracks the complete provenance of an image. The risks associated with these new tools are real, as they could facilitate the creation of deepfakes or copyright infringement. According to Clarity, a deepfake detection company, the creation of these has increased by 900% compared to last year. Additionally, a recent survey by YouGov shows that 85% of Americans are concerned about the spread of misleading deepfakes.

Midjourney will soon let users upload and edit images with AI

Why it matters: Midjourney, the startup behind one of the most popular text-to-image AI models, just announced plans to roll out an upgraded web tool that will allow users to edit any image on the internet using the company's powerful generative AI capabilities. In a message posted to Midjourney's official Discord server last week, CEO David Holtz spilled the details on this upcoming "image editor" feature. Users will be able to upload any pic from across the web. The tool will then use recent advancements in "depth controlnets" to intelligently change an object's shape and textures based on a text description. "It keeps the shape of the scene/object but repaints all the texture/colors/details according to your prompt," reads the description shared by Holtz. Of course, putting such powerful editing abilities into the hands of Midjourney's several million users comes with huge implications around potential misuse. There are always fears that tools like these could be exploited for nefarious purposes like creating misleading deepfakes or violating copyrights on a massive scale. Despite this, the US still lacks a law criminalizing deepfakes at the federal level - though at least a dozen states have enacted statutes against AI impersonation, with many more considering following suit. Holtz seems to recognize the gravity of the situation, saying in his Discord post that Midjourney will initially restrict access to a "subset of the current community" as it works to improve human and AI monitoring systems to prevent abuse of the editor. However, he admits that the company is "not sure" how to implement such restrictions. The company is now polling its user base to get feedback on who should get early access. To their credit, Midjourney has been ahead of the curve on one front by embracing IPTC's Digital Source Type metadata standards that flag AI-generated images. But the company is trailing others, like Adobe, who use more robust provenance such as C2PA, which is another metadata technology that traces an image's full editing history. It's worth mentioning that Midjourney launched a new website in August that lets anyone generate AI images through text prompts by simply signing up with a Google account. Subscription plans are available after an initial free trial period.

Meld your photos with your imagination using Midjourney's new AI image editing tools

Midjourney's AI image creator is adding major new editing options to its platform, as announced by CEO David Holz on Discord. You'll be able to edit an uploaded image using generative AI, as you could only do with images that AI has created until now. Notably, this will be a web-based tool, expanding Midjourney's Discord-centered and exclusive features. While AI editing tools for images have been around for a bit, they are almost always restricted to images created with the same AI model doing the editing. Midjourney is taking the leap into allowing you to edit an image you want to edit rather than having the model make the initial creation. That's technically impressive since it means the model must understand more about an image that it didn't design from scratch. However, it may also be risky since other models may have held back from providing it as an option due to concern over people using it to mess with pictures of real people without their permission or to employ the AI for less-than-benevolent purposes. Once the image is uploaded, you can edit it in several ways. You can zoom in or out, adjust the angle, and otherwise play with the image without needing Discord and with just a single command. You can also use a digital brush for inpainting, allowing for more precise modifications. The most notable new element is the Retexture feature, which lets you precisely alter colors and other details while keeping the image's shapes and objects intact. Questions about the ethics of the creation and editing of AI images have led to heated debates, lawsuits, and other arguments. That makes Midjourney's decisions important for how its rivals and regulators figure out rules to set up internally and for the industry. Midjourney has agreed to embed metadata in AI-generated images so that they can always be identified, though not the more comprehensive C2PA standard as of yet. These are issues that Midjourney is well aware of. "Because these new features are so powerful we're going to restrict the deployment of them to a subset of the current community and increase our human moderation over the outputs (there's also new more advanced AI moderators which we feel will do a great job)," Holz wrote. "Honestly, we're not sure how to precisely restrict deployment of this feature, so we want to put a survey out there and get your opinions. We'll use these results to calculate what percentage of the userbase we'll reach with different restrictions and try to balance accessibility vs volume." For the average user, the ability to easily manipulate images with AI could make using Midjourney more appealing and make photo editing a lot easier. And as AI image makers become more common Midjourney wants to be more accessible. That's why the company decided to abandon Discord as its only outlet, though it's still a major center for the company's creations. Midjourney likely doesn't want to be left out simply because some don't want to sign up for Discord.