Midjourney has solved one of many largest issues with generative AI artwork with the choice to upscale photographs to super-high resolutions.
Till this week, Midjourney’s output was restricted to a default decision of 1,024 x 1,024 pixels. That was massive sufficient for medium-sized net photographs, however not excessive decision sufficient to print at something higher than postcard dimension.
Now the corporate has added the choice to upscale photographs by 2x or 4x, which means the utmost picture dimension is elevated to 4,096 x 4,096 within the default sq. format. In different phrases, that’s a 4K decision picture.
The upscaler will be utilized to pictures with completely different facet ratios, too. On a 16:9 picture, for instance, the 4x upscaling delivers a decision of 5,824 x 3,264.
This large bounce in decision means Midjourney’s AI-generated photographs might be printed at A3 dimension and never look blurry or pixelated. The decision can also be greater than adequate for creating high-quality net photographs or desktop wallpapers, for instance.
The Value Of Upscaling
Upscaling photographs to those new resolutions does come at a price. While upscaling to the default picture dimension usually solely takes just a few seconds, 4x upscaling jobs can take two or three minutes to finish.
Midjourney prices prospects for the quantity of processing time they eat, so upscaling photographs will eat into the variety of photographs prospects can generate inside their month-to-month allowance. Clients will pay additional to extend the variety of hours they’re permitted, nevertheless.
The corporate additionally warns that the upscalers could also be temperamental, advising that in “very uncommon” cases upscaling might lead to “black or corrupted picture on 4x upscales”. The corporate additionally provides that “blurry” photographs gained’t be improved by upscaling and that photographs would possibly get barely darker when upscaled.
Midjourney’s new upscalers have been launched only a week after Adobe raised the decision bar for AI-generated photographs with the launch of Firefly 2.
Adobe bumped the decision of its Firefly photographs to 2,048 x 2,048—the identical decision as Midjourney’s 2x upscaled photographs, however solely half the decision of the 4x upscaler.
It’s indicative of a rising arms race within the generative AI artwork business, with corporations making an attempt to outdo each other on picture high quality and determination. Open AI lately introduced DALLE 3, which—along with claimed enhancements in its skill to ship photographs described in textual content prompts—additionally bumped the default decision to 1,729 x 1,024.