AI News: Midjourney V8 vs Microsoft MAI-Image-2 Compared

TL;DR

Two new image models launched this week, and Microsoft's MAI-Image-2 outperformed Midjourney V8 in testing. MAI-Image-2 nailed detailed text like coffee shop menus and photorealistic scenes, while Midjourney V8 struggled with instruction-following and mangled hands, though it still excelled at surreal, imaginative prompts. MAI-Image-2 ranks third on the text-to-image arena behind OpenAI and Nano Banana.

Transcript

Once again, there's been a ton of news in the world of AI, and I want to help you separate the signal from the noise and tell you what I think you should know, and also just some of the more interesting and cool stuff that happened. I'm not going to waste your time. Let's jump right into it. Let's start with a company that I haven't talked about in... Read More

Key Insights

• Midjourney V8 was released with improvements in following detailed directions, aesthetic personalization, style references, image coherence, and text rendering, plus a new HD mode that renders images in 2K resolution and a redesigned web interface.
• Midjourney V8 is extremely fast, noticeably faster than version 7, but the online consensus and the reviewer's own tests found its instruction-following and photorealism underwhelming compared to competitors like Nano Banana.
• Midjourney V8 still produces mangled anatomy, with test images showing broken arms, three-fingered hands, and duplicate limbs, despite the expectation that AI models had solved the hand-rendering problem a while ago.
• Midjourney excels at surreal, weird, and imaginative prompts, figuring out interesting interpretations of random ideas, but performs worse when asked to follow specific, detailed instructions.
• Microsoft introduced MAI-Image-2 this week, an unexpected release currently ranked the third-best image model on the text-to-image arena behind OpenAI's model and Nano Banana.
• MAI-Image-2 focuses on photorealism, built for creatives wanting images with natural light, accurate skin tones, and lived-in environments, plus reliable in-image text generation and rich detailed scenes.
• MAI-Image-2 accurately rendered a full coffee shop menu board, including item names, prices, hours, and a Wi-Fi password, demonstrating text-generation quality comparable to Nano Banana.
• The video also covered other announcements referenced in the description, including Google Stitch AI, Google AI Studio vibe coding, NVIDIA DLSS 5, and OpenAI's GPT-5.4 Mini and Nano models.

Questions & Answers

Q: What is new in Midjourney V8?

Midjourney V8 is the eighth version of the model, improved at following detailed directions, understanding your aesthetics through personalization and style references, and producing more coherent, detailed images. Text rendering works better than before, historically Midjourney's weakest area. It retains old parameters like chaos, weird, and raw, adds an upgraded web interface, and includes a new HD mode that renders images in 2K resolution.

Q: Is Midjourney V8 better than Nano Banana for image generation?

No, based on the reviewer's testing and online consensus, Midjourney V8 falls short of Nano Banana. In side-by-side comparisons, Midjourney produced worse results, including mangled hands and anatomy, while Nano Banana handled the same prompts more coherently. The host noted Midjourney may have once been the state-of-the-art image model, but tools like Nano Banana now feel far ahead of it in quality and instruction-following.

Q: How fast is Midjourney V8 compared to version 7?

Midjourney V8 is extremely fast, way faster than version 7 according to the reviewer's hands-on testing. This speed was one of the most noticeable improvements during his tests. However, the increased speed did not translate into better quality, as the model still struggled with instruction-following, image scaling, and text rendering, producing some results the host described as looking like Midjourney V1 or V2 level.

Q: Why does Midjourney still generate bad hands and anatomy?

Despite expectations that AI models fixed hands a while ago, Midjourney V8 still produces distorted anatomy. Test images showed a woman with only three fingers, broken arms, and duplicate limbs reaching the same microphone. Multiple reviewers, including Alex Petrescu and Jav Lopez, shared examples where zooming into hands revealed mangled results, confirming the problem persists in V8 even though it was thought to be solved.

Q: What is Microsoft MAI-Image-2?

MAI-Image-2 is an image generation model Microsoft introduced this week, an unexpected release. It currently ranks as the third-best image model on the text-to-image arena, behind OpenAI's image model and Nano Banana. It focuses on photorealism, built for creatives who want images with natural light, accurate skin tones, and lived-in environments, plus reliable in-image text generation and rich, detailed scene generation.

Q: How good is MAI-Image-2 at generating text within images?

MAI-Image-2 handles in-image text very reliably, at a level the reviewer compared to Nano Banana. When tested on a coffee shop menu board, it accurately rendered every detail: Orbit Cafe, espresso at $3, cappuccino at $4.50, vanilla latte at $5, cold brew at $4, avocado toast at $7, hours of 7 a.m. to 4 p.m., and the Wi-Fi password 'stay curious.' The host said it nailed the task exactly as intended.

Q: Which is better, Midjourney V8 or Microsoft MAI-Image-2?

In the reviewer's opinion, Microsoft's MAI-Image-2 is the more impressive of the two image models released this week. MAI-Image-2 followed prompts well and produced convincing photorealism, accurate text, and coherent surreal images like a transparent glass sneaker containing a miniature ocean. Midjourney V8, by contrast, botched scaling, produced blurry text, and mangled anatomy, though it still excelled at weird, imaginative prompts over specific, detailed ones.

Q: What other AI announcements happened this week?

Beyond the two image models, the video's resources reference several announcements: Google Stitch AI for UI design, full-stack vibe coding in Google AI Studio, Google AI Mode personal intelligence expansion, NVIDIA NemoClaw, NVIDIA DLSS 5, NVIDIA Space Computing, and OpenAI's GPT-5.4 Mini and Nano. The host framed the week as another packed period of AI news, aiming to separate the signal from the noise for viewers.

Summary & Key Takeaways

• Midjourney released V8, promising better instruction-following, personalization, style references, coherence, and improved text rendering, along with a new HD mode rendering at 2K resolution. Despite being much faster than V7, both online reviewers and the host's own tests found the output underwhelming, with mangled hands and poor photorealism versus rivals.

• The host tested Midjourney V8 on instruction-following, text rendering, and a surreal glass-elephant prompt. Results were mixed: some images botched scale and details, text came out blurry and mangled, but the surreal prompt showed Midjourney still excels at weird, imaginative compositions rather than precise, specific requests.

• Microsoft unexpectedly launched MAI-Image-2, ranked third on the text-to-image arena. It focuses on photorealism and reliable text. In tests it accurately rendered a rainforest scene, a full coffee shop menu with prices, and a surreal transparent glass sneaker, proving more impressive than Midjourney V8 overall.