Midjourney's latest release, version 5.2, astounded AI artists with its camera-like feature that allows for zooming out while maintaining the central synthesized image. This new feature expands the borders of an existing AI-generated image and creates a larger scene around it, simulating the effect of zooming out with a camera lens.
Similar to OpenAI's DALL-E 2 technique called outpainting, introduced in August 2022, Midjourney's zoom-out feature keeps the original subject centered while enlarging the image. However, unlike DALL-E and Photoshop's Generative Fill, this feature currently only works on images generated within the Midjourney platform, a subscription-based AI image-generator service.
On the Midjourney Discord server, where users can experiment with the platform's features, the zoom-out capability is accessible by generating a v5.2 image and upscaling the result. Additional buttons appear below the output, enabling users to zoom out by factors of 1.5x, 2x, or a custom value between 1 and 2. Another button, labeled "Make Square," generates material around the existing image, creating a 1:1 square aspect ratio.
David Holz, the creator of Midjourney, announced the new features and improvements of v5.2 on the Discord server. Besides the zoom-out functionality, significant additions include an enhanced aesthetic system for better image quality, a stronger "--stylize" command for influencing the level of realism, an activated "high variation mode" for increased compositional variety, and a new "/shorten" command to trim non-essential words from prompts.
Holz emphasized that changes to the platform might occur without notice, even though v5.2 was immediately rolled out. Users still have the option to access older versions of the Midjourney model using the "/settings" command or the "--v 5.1" in-line command argument.
Enthusiasts of this emerging art form, known as "synthography" by some proponents, expressed their admiration for the v5.2 updates, using terms like "stunning" and "mindblowing." While the hype around AI is often accompanied by such superlatives, fans argue that the visual improvements in Midjourney justify these astonished reactions.
This latest update is part of a series of quality enhancements since March 2022, when the model initially generated images with limited definition and detail. Previous releases, v5.0 in March and v5.1 in May of the same year, improved realism and image detail. Notably, the v5 model series allowed the creation of realistic images of Pope Francis and Donald Trump, raising concerns about deepfakes on social media.
Despite the excitement among Midjourney enthusiasts, image synthesis remains a controversial topic for some artists due to the ethical concerns surrounding the training of AI systems. These systems often utilize millions of scraped images from the web without artist consultation, credit, or permission. The exact contents of Midjourney's training data have not been officially disclosed. While Adobe's Firefly aims for a more ethical approach, it was reported that active artist consent remains limited.
While appreciating Midjourney's impressive technical advancements, it is essential to consider a more ethical path forward for this technology, one that addresses the concerns of both traditional and synthography artists.
Leave a Comment