How much detail is too much? Midjourney v6 attempts to find out

0
117

[ad_1]

Enlarge / An AI-generated picture of a “Lovely queen of the universe wanting on the digital camera in sci-fi armor, snow and particles flowing, fireplace within the background” created utilizing alpha Midjourney v6.

Midjourney

In December, simply earlier than Christmas, Midjourney launched an alpha model of its newest picture synthesis mannequin, Midjourney v6. Over winter break, Midjourney followers put the brand new AI mannequin via its paces, with the outcomes shared on social media. Up to now, followers have famous far more element than v5.2 (the present default) and a special method to prompting. Model 6 can even deal with producing textual content in a rudimentary approach, however it’s removed from good.

“It is positively a loopy replace, each in good and fewer good methods,” artist Julie Wieland, who incessantly shares her Midjourney creations on-line, instructed Ars. “The small print and surroundings are INSANE, the draw back (for now) are that the generations are very excessive distinction and overly saturated (imo). Plus it’s essential sort of re-adapt and rethink your prompts, working with new constructions and now much less is sort of extra when it comes to prompting.”

On the similar time, critics of the service still bristle about Midjourney coaching its fashions utilizing human-made paintings scraped from the net and obtained with out permission—a controversial observe frequent amongst AI mannequin trainers we have now covered in detail up to now. We have additionally lined the challenges artists may face sooner or later from these applied sciences elsewhere.

An excessive amount of element?

With AI-generated element ramping up dramatically between main Midjourney variations, one may marvel if there may be ever akin to factor as “an excessive amount of element” in an AI-generated picture. Midjourney v6 appears to be testing that very query, creating many pictures that typically appear extra detailed than actuality in an unrealistic approach, though that may be modified with cautious prompting.

In our testing of model 6 (which may at present be invoked with the “–v 6.0” argument on the finish of a immediate), we seen occasions when the brand new mannequin appeared to supply worse outcomes than v5.2, however Midjourney veterans like Wieland inform Ars that these variations are largely because of the totally different approach that v6.0 interprets prompts. That’s one thing Midjourney is repeatedly updating over time. “Previous prompts typically work a bit higher than the day they launched it,” Wieland instructed us.

[ad_2]

Source link