Midjourney V6 With Text Generation Is Here—See How Dramatically It Compares
By Mikelle Leow, 22 Dec 2023
Image generated on Midjourney V6
“Midjourney V6 is just real life,” one user revels. The latest model by the popular text-to-image artificial intelligence generator touts a much-improved image quality, serving up visuals that are more lifelike and intricately detailed than ever before.
The updated version can understand prompts much better, as well as makes room for more extended input. As such, users can expect more coherent responses and an expanded knowledge base from the app.
Here’s what churned out when we entered the description, “a gallery with frames of blurred faces, lit by low lighting,” in both versions 5.2, the closest predecessor, and 6.0.
Prompt: a gallery with frames of blurred faces, lit by low lighting
Images generated on Midjourney V5.2 (on the left) VS V6 (on the right)
One of the most exciting additions is its newfound ability to generate legible text within images, a challenging feat that earlier versions of Midjourney struggled to achieve.
Prompt: photo of a candy store with an ominous glow in the middle of the woods. On its facade spells the words “Free Candy”
Images generated on Midjourney V5.2 (on the left) VS V6 (on the right)
Prompt: closeup photo of a pink frosted cake with words in green icing spelling “You Tried”, cake is on a table with a tablecloth
Images generated on Midjourney V5.2 (on the left) VS V6 (on the right)
By extension, Midjourney V6 recognizes company logos and can accurately depict them in its outputs, making it far more possible to dream up branding concepts.
Midjourney v6 excels at generating trademarked logos, I'll give it that. pic.twitter.com/I6Ed4xAxcv
— Howard Pinsky (@Pinsky) December 21, 2023
5. Prompt: A Coca Cola ad, featuring a beverage can design with traditional Hawaiian patterns pic.twitter.com/nqow7UngX4
— Chase Lean (@chaseleantj) December 21, 2023
The development journey of V6 spanned nine months, marking it as the third model built from the ground up using Midjourney’s AI superclusters.
The AI doesn’t get it right all the time, but it’s still more advanced by a long shot when it comes to interpreting text in both inputs and outputs.
Trying out the latest Midjourney on your own
Midjourney V6 is currently in alpha version, so it won’t be selected by default. To access it, you will either have to fire up the settings menu by typing “/settings” in the textbox or use the command “--v 6” after your prompts.
However, the leap from V5 to V6 isn’t without its learning curve. Users familiar with previous versions will need to adapt to V6’s heightened sensitivity to prompts. As detailed by Midjourney founder David Holz in a Discord server, it will hurt your experience if you continue to use “junk” terms like “award winning, photorealistic, 4k, 8k,” to guide the AI, despite those keywords being useful in the past.
Users should strive to be more descriptive to attain the desired outcomes on the V6 model. “Be explicit about what you want,” Holz reiterates. “It may be less vibey, but if you are explicit, it’s now MUCH better at understanding you.”
For images with text in them, users should wrap the words in “quotations” when entering them in the text field.
For example, to create higher-quality pictures of a protest with signs that were actually readable, we used the command, “photo of activists at a protest, with one sign reading ‘Spill the Tea’ --v 6.”
Prompt: photo of activists at a protest, with one sign reading “Spill the Tea”
Images generated on Midjourney V5.2 (on the left) VS V6 (on the right)
The update also introduces improved upscalers, featuring both ‘subtle’ and ‘creative’ modes, to enhance the resolution of generated images.
You can also make adjustments to how realistic or creative you want the machine to be in interpreting your descriptors.
“If you want something more photographic / less opinionated / more literal, you should probably default to using --style raw,” notes Holz. “Lower values of --stylize (default 100) may have better prompt understanding while higher values (up to 1000) may have better aesthetics.”
The proof is in the /pudding
Early adopters are already singing praises for V6’s photorealism and detail. We’ll just let the images do the talking from here on out.
Midjourney
— Nick St. Pierre (@nickfloats) December 21, 2023
Dec 2022 (v4): Dec 2023 (v6): pic.twitter.com/Q2WpvZK4xN
2. Prompt: A realistic standup pouch product photo mockup decorated with bananas, raisins and apples with the words "ORGANIC SNACKS" featured prominently pic.twitter.com/M3IIMeyHpJ
— Chase Lean (@chaseleantj) December 21, 2023
4. Prompt: A magazine quality shot of a delicious salmon steak, with rosemary and tomatoes, and a cozy atmosphere pic.twitter.com/CQls0mUTbE
— Chase Lean (@chaseleantj) December 21, 2023
Just wrapped up a photoshoot...
— BLΛC (@blac_ai) December 22, 2023
In Midjourney v6. pic.twitter.com/3milaxSTIe
Big performance boost from #midjourneyV6 . Love how fast these tools are evolving! Been closely following people's experiments.
— Andrey Seas (@andrey_seas) December 22, 2023
Have you experimented yet? What are your thoughts? pic.twitter.com/FaqIsSy5Kp
Good morning.
— Brian ð¨ Bri Guy AI (@bri_guy_ai) December 21, 2023
Are we enjoying v6?
Here’s an early favorite from Midjourney’s latest version. pic.twitter.com/tZNBqjrad5
ãã®1å¹´4ã¶æã¡ããã§ã®çæAIã¹ã·ã®é²å
— ç«é»å¤èº¯ (@NatsukuPhoto) December 21, 2023
← æ¨å¹´8æï¼Mid V1ï¼ ä»æ¥ ( #midjourney V6 )→ pic.twitter.com/neTiksGbBu
[via VentureBeat, Slashdot, Gizmodo, images via various sources]