Midjourney:自然语言生成图片的AI的魅力与局限
“Midjourney是一款近期备受关注的AI图像生成工具,其 ability to create images based on natural language descriptions has garnered significant attention online. After a brief trial, I found it to be a fascinating tool and decided to purchase a $30 per month subscription for unlimited usage. Using this technology, I was able to generate several images, particularly those related to architecture, as it is my area of expertise. Additionally, many of the images I created were inspired by existing masterpieces, serving as a nod to the creative spirit of the time.”
据我推测,Midjourney 的机器学习素材来源应以丰富的艺术和摄影图库为主。相较于其他 AI 技术如 Standard Diffusion 或 Dalle 2,Midjourney 生成的作品虽然在某些细节上可能不够完美,但至少在构图、色调和光影方面表现出色,给人一种整体舒适的感觉。偶尔运气好的时候,甚至能生成具有现实设计指导意义的优秀作品。
图片下的文字即为生成所用的 prompt,大家有兴趣可以自己试试。
cyberpunk style cathedral with skylight
modern buddhist temple designed by tadao ando, wood structure, cast in place concrete, skylight, photo realistic
front view of a brutalist structure with lush greenery, soft light
posthumanism architecture
kengo kuma designed a meditative bamboo temple space with people sitting, chanting that showcase the connection joints of the columns
white modern style villa on a sea side cliff, designed by Richard Meier, forest and snowy mountain in the background
monet painted manhattan skyline
peter zumthor chapel that has a dramatic skylight on top made out of concrete and wood
peter zumthor chapel interior made with rammed earth, that has skylights
steampunk city in ancient chinese style, view from sea, pagotas, mountains in background, raining, volumetric lighting,
shigeru Ban designed meditative temple space
a biomorphic greenhouse
futuristic city with a lot of plants , high spead train, dense buildings and water
several floating steampunk city islands on the ocean
aerial view of a seaside cyberpunk city, mountain background on left side, lots of boats and ships on the sea, cinematic lighting
当我深入了解Midjourney时,其强大的能力给我留下了深刻的印象,尽管它并非立即就能替代建筑师的工作。我认识到,工具本身并不能创造价值,只有那些富有智慧的创造者才能利用工具使自己变得更加强大。然而,在这里,我所体验到的震撼是:我越来越难以区分机器学习和人类智慧之间的本质差异。
虽然当前AI生成的图像存在一些不尽如人意之处,其中的细节并不十分清晰,这主要源于AI学习的图库主要是基于二维图片,尚不能充分理解图片中色彩线条所代表的物理现象。因此,AI生成的结果在本质上其实只是一堆颜色。然而,随着科技的不断发展,我们会看到越来越多的真实数据被融入到算法之中,例如各种材料的物理属性、物理定律等。在这样的背景下,未来的AI将能够生成更为精确和逼真的图像,其创作能力或许将会与人类的直接创作相媲美。
AI时代,拥有个人微信机器人AI助手!AI时代不落人后!
免费ChatGPT问答,办公、写作、生活好得力助手!
搜索微信号aigc666aigc999或上边扫码,即可拥有个人AI助手!