Midjourney's development roadmap over the next few months will focus on two major improvements to image generation: style personalization through "personality tuning" tests and enhanced upscaling. According to David Holz, Midjourney's CTO, these features are currently in the testing and tuning phase and could be released within weeks. Beyond aesthetic upgrades, they represent a shift towards more user control over the AI assistant's artistic style and technical capabilities.
Customized Style Through Personality Tests
One of the most exciting upcoming features is the ability to create custom "personalities" for Midjourney using specialized tests. Currently, Midjourney has a single style that manifests across all prompts. With customizable personalities, users could potentially tune Midjourney for specific projects to generate images that better match a desired aesthetic.
For example, a fantasy author might create a Midjourney personality focused on dark, gritty illustrations. An architect could develop a personality optimized for sleek 3D renderings. Holz explained that these personalities may or may not be shareable with others. Either way, it opens creative doors by expanding Midjourney's artistic range beyond what is possible through prompts alone.
Sharper Images Through State-of-the-Art Upscaling
In tandem with style personalization, Midjourney aims to roll out more advanced upscaling. While the current model can enlarge images up to 1024x1024 pixels, new upscalers in testing can reach resolutions as high as 3000x3000 pixels while preserving fidelity.
Midjourney is experimenting with two different upscaling approaches. One takes a more conservative approach to minimize artifacts and distortions. The other uses a more holistic process that resynthesizes the entire image to fix flaws and maximize detail. The result will be larger images with superior quality, revealing intricacies previously lost at lower resolutions.
A Vision into Midjourney's Future
With these updates, Midjourney continues to rapidly evolve month-to-month. Looking further ahead, even more transformative changes are on the horizon. Holz hinted at 3D image generation in the coming months, with video synthesis not far behind. The subscription revenue from Midjourney's 1 million+ users will fund expanded AI research and computing power to make these advances possible.
No Mention of DALL-E v3
Notably absent from the discussion was any mention of rising competition from platforms like DALL-E. OpenAI recently unveiled DALL-E 3, which introduces capabilities like text-to-image generation that many believe could pose a threat to Midjourney's dominance. DALL-E 3 also features tight integration with ChatGPT for more natural language prompting.
With impressive samples spreading across social media, some are hailing DALL-E 3 as a "Midjourney killer." While Midjourney maintains a solid user base, ignoring parallel progress from other generative AI providers risks complacency. To retain its position at the forefront of the space, Midjourney will need to continue aggressively improving its core features while pushing into new artistic frontiers.
While the launch timeframe for new capabilities remains fluid, Midjourney's commitment to pushing creative boundaries is clear. Personality tuning and upscaling exemplify how the platform places control and customization in the hands of users. Midjourney's openness about its internal roadmap also fosters community transparency as the technology advances. For artists and AI enthusiasts alike, it is an exciting phase as Midjourney develops alongside its users.