AI-powered art generator Midjourney hasn’t enjoyed the same coverage as viral chatbot ChatGPT, but it’s certainly found favor with creatives keen to leverage its features to generate unique images from natural language inputs. The latest release, Version 5, coincides with the launch of GPT-4 as AI remains by far the hottest topic in tech this year.
Announced on March 15, Midjourney V5 is said to be a major step change on its predecessor, boasting “much higher image quality, more diverse outputs, wider stylistic range, support for seamless textures, wider aspect ratios, better image prompting, wider dynamic range and more.”
For the uninitiated, Midjourney uses AI to generate images from text prompts, similar to OpenAI’s DALL-E and Stable Diffusion.
Users put Midjourney V5 through its paces
The announcement that Midjourney V5 was available for community testing was met with a flurry of responses as users wasted no time in comparing the release to the previous version.
In one thread, designer Nick St. Pierre provided a series of side-by-side images comparing the output of V5 to the shiny new model, posting photos of living rooms, fashionable 1960s females, snakeskin Nike sneakers, and sci-fi movie scenes.
Although in some cases the results don’t appear to be hugely different, St. Pierre noted that he had not changed the prompt for his exercise – and that “V5 prompting is going to be slightly different from V5.”
The designer subsequently produced another thread detailing the ways in which he had adapted his prompting for the new version to include more natural language, noting that “It’s working and freaking me out how good the results are.”
After putting V5 through its paces, writer Jeff Hayward said that “each result drew a silent ‘wow’ from me as I sit alone in my office. From people to movie monsters to architecture, V5 creates more realistic images than previous versions without complicated prompts.”
Although it’s only been four months since Midjourney Version 4 was released, the latest iteration is notable for its superior level of detail, depth and accuracy, although some familiar problems remain – notably AI’s apparent inability to accurately render human hands. Multiple users said the tool returned a hand with six fingers when given a simple prompt to produce an image of a photo realistic hand.
It seems like only a matter of time before this pain point is resolved. Back in August, Midjourney founder David S Holz said, “Whatever seems good today, will seem irrelevant in 30 to 60 days… I hope you all come along for the ride.”
Midjourney magazine to launch
Midjourney is available via its Discord channel, and users will need to adjust their settings to access V5. This can be done by running the prompt /settings and selecting the latest version, which is helpfully situated next to version 4.
Perhaps unsurprisingly given the improved quality, Midjourney V5 is slower at processing images than V4 – although many will see this as a small price to pay for access to almost unlimited photorealistic imagery.
Interestingly, Midjourney made another major announcement on March 15 with the revelation that it would launch a monthly magazine.
It seems the publication will act as a showcase for the tool, with the bulk of the contents being “breathtaking images and prompts” – although there will also be interviews, likely with those who have successfully used Midjourney in their professional lives. Users who subscribe get the first issue for free.
This article is originally from MetaNews.