Midjourney思维模式测试结果
有人测试了Midjourney新思维模式，结果发现普通模式反而更美观，但思维模式在文本准确度上有优势。
用户对Midjourney的思维模式进行了快速测试，比较了三种图像生成结果。GPT图像获得8.0/10分，场景构图和文本准确度最高。Midjourney思维模式得5.3/10分，氛围和光照出色但细节不足。普通Midjourney模式得5.0/10分，环境细节好但广告准确度较低。从美观度看，普通Midjourney模式表现最佳。
I gave @Midjourney's thinking model a *very* quick test. Prompt is in the first comment, link to the chat with all the ratings in the second. Results are below:
Image 1: 8.0/10 – Excellent scene composition, crowd density and accurate text. Minor omissions and unwanted lettering.
Image 2: 5.3/10 – Strong atmosphere and lighting, but insufficient pedestrians, missing details and several text errors.
Image 3: 5.0/10 – Good environmental detail and reflections, but weaker advertising accuracy, sparse crowds and incorrect signage.
Verdict: Image 1 wins comfortably. Images 2 and 3 are broadly comparable, with Image 2 narrowly ahead on exact text accuracy.
Images are:
1. GPT Image 2. Midjourney thinking mode 3. Midjourney without thinking mode
Midjourney settings were vanilla, no personalisation, and I have to say, ignoring accuracy, Midjourney comes out looking a lot nicer imho.
Note: I'm not saying this is the point of thinking mode. This was just an experiment.
