AI Deepfakes Just Got Better With This Upgrade

MidJourney was already one of the most impressive AI image generators out there, and with the release of version 5.1, it’s now become much more powerful. The algorithm update means that, even for beginners, it’s now incredibly easy to create stunning works of art with minimal effort—and deepfakes. Lots and lots of deepfakes.

The AI image generator entered the public consciousness last month after deepfake images of Pope Francis in a white puffer coat went viral. The controversy sparked conversations about the ethics of AI generators like MidJourney, and the company took action soon after.

Use of the word «pope» on MidJourney is now banned, but creating deepfakes of other people, such as President Joe Biden, is still allowed—probably until we see a viral image of Biden wearing a pink neon spandex jumpsuit doing yoga in the Oval Office.

Or maybe one of Biden wearing an intergalactic spartan costume that looks kinda like Iron Man.

Joe Biden as an intergalactic spartan, created with MidJourney.

MidJourney also allows users to create high-quality deepfakes using photos as a reference. This feature has significantly increased MidJourney’s popularity due to the impressive results it provides and its low learning curve. At the moment, it is the most popular AI image model in terms of user interest, surpassing Dall-e and Stable Diffusion.

The latest upgrade to MidJourney promises improvements in the way the algorithm interprets users’ prompts. Version 5.0 was released just three months ago and marked a significant departure from Version 4.

V5.1 is now available! Images are more coherent, sharp, and beautiful. It’s easier to use and should respond more precisely to instructions. We’ve also added a «RAW» mode for expert users to reduce the ‘opinionatedness’ of our model and give you more creative control.

— Midjourney (@midjourney) May 3, 2023

This new model offers higher coherence, more accuracy to text prompts, fewer unwanted borders or text artifacts (the weird random things that sometimes appear on images), and sharper results overall, according to MidJourney. The icing on the cake is that the new version was trained to be more «opinionated.»

Being opinionated means that the model has the freedom to be more creative when interpreting a user’s prompt. This results in better images with basic prompts—even with absurd commands of just one or two words.

For example, simply comparing the results for «Bitcoin Wallet» prompts using version 5.1 and 5.0 demonstrates the differences and improvements. Version 5.1 imagines a leather wallet, with a bitcoin logo embroidered and placed on a table. Version 5 simply shows a Wallet with a Bitcoin on a plain surface.

The new version further emphasizes the strengths that have led to MidJourney’s explosion, without forgetting experienced users who now have a RAW mode if they want more accurate and less «opinionated» results.

And what about the many-fingered hands? MidJourney v5.1 has greatly improved. Hand representations are now successful in most cases, although the generator still produces some errors.

The MidJourney team says we might see version 5.2 in the future, further enhancing the capabilities of this tool. If third time’s the charm, then the days of AI generators messing up hands in our favorite deepfakes may finally be behind us.

Image created by Decrypt using AI (MidJourney V5.1)

Source