Midjourney’s new V6 can generate meaningful text

Midjourney has announced a new AI-powered image-generation model that can generate meaningful text.

We’re now alpha-testing our V6 models Midjourney. Just type /settings and click V6 or add —v 6 after your prompt. Image coherence and prompt understanding are greatly improved. You can draw text and dolphins and there’s new upscalers too. Happy holidays everyone!

— Midjourney (@midjourney) December 21, 2023

According to a post on the Discord channel, Midjourney V6 is positioned as a \”ground-up overhaul\” and vastly surpasses the predecessor released in March this year.

“More precise adherence to guidelines, as well as longer prompts, improved coherence and model knowledge,” the developers described the new version of the neural network.

One of the most important components of V6 is its ability to generate meaningful text rather than a string of characters as before. However, the developers noted that this is not the central element of the model.

Thanks to the improvements, Midjourney can now compete with leading generative models such as DALL-E 3 and Ideogram.

Due to the new capabilities, the sixth version of the AI has become somewhat slower and more expensive, but the developers have promised to work on its performance in the future.

Midjourney V6, trained from scratch on AI superclusters, also boasts improved scaling modules in \”fine\” and \”creative\” modes, doubling image resolution.

The alpha version of the neural network supports several new arguments, including \”-ar\” for changing resolution, \”-chaos\” for selecting image generation variations, \”-stylize\” for changing the creative style (the smaller the value, the faster the network will produce a result, at the cost of detail).

“You’ll have to relearn how to prompt,” the project team warned.

Other features such as zoom and panning will appear during the model’s release.

For now, you can participate in testing Midjourney V6 on the Discord server by sending the desired prompt in the chat.

“The project team has implemented moderation systems and will ensure compliance with our community standards with heightened strictness. Do not be idiots and do not create images that provoke drama,” the developers stressed.

Users have already started sharing the results.

Midjourney V6 …. We have TEXT!
It can be hit or miss but still learning how it works.
These 4 were all from one generation.
Maybe got lucky ?‍♂️
Prompt in image 1 ALT#midjourneyV6 #MJV6 #AIart #aiartcommunity #digitalart #midjouney #MidjourneyAI #AIArtwork pic.twitter.com/BIJwr49489

— Orcton (@OrctonAI) December 21, 2023

Most note a high degree of detail in the generated images even without upscaling.

Midjourney V6 is looking mighty tasty!
Huge increase in detail at the same resolution.
These are NOT final model images and are not upscaled.#midjourney #AIArtCommuity pic.twitter.com/SJYK6v7LyC

— GifCo (@giffboake) December 20, 2023

To showcase Midjourney’s capabilities, one user asked to generate “a 92-year-old wrinkled angry old man in studio lighting.”

Midjourney v1 until v6, same prompt”

white background, closeup portrait of a very old mean man, 92 years old, wrinkles, realistic skin, studio lighting,, canon f/4#midjourneyV6 #midjouney #aiartcommunity pic.twitter.com/g8wAALAbH3

— Marco Nedermeijer (@MNedermeijer) December 21, 2023

In November, the AI startup Anthropic announced a new model Claude 2.1 with an expanded context window of 200 000 tokens, equivalent to 500 pages of text. 

Earlier, Google announced a global update for the Bard chatbot intended to extend the capabilities of the neural network. The update is named Gemini.

In November, the AI startup Anthropic announced a new model Claude 2.1 with an expanded context window of 200 000 tokens, equivalent to 500 pages of text.