“Amazing” – Midjourney update stuns AI performers with camera-like feature

On Thursday, Midjourney unveiled version 5.2 of its AI-powered image synthesis model, which includes a new “zoom out” feature that preserves a centrally synthesized image while automatically building a larger scene around it, simulating zooming out with a camera lens.

Similar to outpainting – an AI imaging technique introduced by OpenAI’s DALL-E 2 in August 2022 – Midjourney’s zoom-out feature can take an existing AI-generated image and magnify its boundaries, while keeping the original subject centered in the new image. But unlike DALL-E and Photoshop’s Generative Fill feature, you can’t select a custom image to expand. At this time, the v5.2 zoom out only works on images generated within Midjourney, a subscription-based AI image generator service.

On the Midjourney Discord server (still the official interface for Midjourney, although there are plans to change that), users can experiment with zooming out by generating a v5.2 image (now the default) and scaling up a result . After that, special “Zoom” buttons appear below the output. You can zoom out by a factor of 1.5x, 2x, or a custom value between 1 and 2. Another button, called “Square”, generates material around the existing image in a way that creates a 1:1 square aspect ratio.

David Holz, the creator of Midjourney, announced the new v5.2 features and improvements on the Discord server on Thursday night. Aside from “zoom out”, the main additions include an overhauled aesthetic system, which promises better image quality and a stronger “–stylize” command that effectively affects how unrealistic an image looks. There’s also a new “high variation mode”, activated by default, which increases compositional variety between image generations. In addition, a new “/shorten” command allows users to rate prompts in an attempt to cut out non-essential words.

Despite the immediate rollout of v5.2, Holz emphasized in his announcement that changes may occur without notice. Older versions of the Midjourney model are still available using the “/settings” command or the “–v 5.1” in-line command argument.

For fans of this new form of image synthesis, sometimes referred to as “synthography” by proponents such as Julie Wieland, the changes in v5.2 are welcome, with some Midjourney users calling them “astonishing” And “mind blow“Which aren’t uncommon superlatives at this point in the hype-friendly world of AI. But fans will likely argue that Midjourney’s visual enhancements are the astounding reactions among themselves.

The latest update is part of a series of quality improvements since March 2022, when the model generated relatively poorly defined images with no detail. Most recently, Midjourney introduced v5.0 in March and v5.1 in May of this year, both of which improved realism and image detail. The introduction of the v5 model series allowed realistic depictions of Pope Francis and Donald Trump to be created, leading to concerns about deepfakes on social media.

Despite the excitement over the new features among Midjourney enthusiasts, image synthesis remains highly controversial among some artists due to the way these AI systems are trained, using millions of images scraped from the internet without consulting the artist, credit or permission. Midjourney has never officially revealed the exact contents of his training data. Adobe is trying to take a more ethical path with Firefly, but Venture Beat recently reported that active artist consent is still marginal.

For now, it’s hard not to appreciate Midjourney’s dazzling technical advancements, while still wondering if there’s a more ethical way forward for this technology – one that satisfies artists and synthographers alike.