关于深度学习:将ChatGPT变成Midjourney提示生成器

101次阅读

共计 4060 个字符,预计需要花费 11 分钟才能阅读完成。

曾经有人总结过能够让 ChatGPT 作为 Midjourney 图像生成的模板。在本文中,咱们将展现如何依据集体用例创立这些提醒,这能够让 ChatGPT 生成的提醒可控性更高。

抉择提醒模板

Midjourney 的模板结构性很强,所以咱们这里应用一下的构造:

让 ChatGPT 为咱们工作

为了让 ChatGPT 工作,咱们还要创立 ChatGPT 的提醒,这样能力让它帮咱们生成 Midjourney 的提醒。

ChatGPT 的提醒的作用是让 ChatGPT 迭代咱们的想法并创立与之相干的图像提醒。咱们只对 ChatGPT 提出咱们的想法和要求,让 ChatGPT 为咱们工作。

一个牢靠而无效的 ChatGPT 提示符有两个次要成分:指令和上下文

首先咱们从上下文开始:咱们能够简略地将咱们的要求变为前缀、场景和后缀:

 #这是 Midjourney 的命令
 
 /imagine prompt: PREFIX, SCENE, SUFFIX
 
 PREFIX defines the image's medium and style
 SCENE defines the content
 SUFFIX modulates PREFIX and SCENE

在下一步中,就是要将咱们将上下文与 ChatGPT 如何解决信息的指令联合起来:

 This is the basic prompt anatomy for image generation with Midjourney:
 
 /imagine prompt: PREFIX, SCENE, SUFFIX
 
 PREFIX defines the image's medium and style
 SCENE defines the content
 SUFFIX modulates PREFIX and SCENE
 
 Create 10 prompts that illustrate the theme "father-son relationship"

下面和上面的限定就是咱们对 ChatGPT 的要求,ChatGPT 会依据咱们的要求来返回后果,大略是这样的:

 1. /imagine prompt: Vintage sepia photograph, A father teaching his son how to ride a bicycle in a park, With a sense of warmth and nostalgia.
 
 2. /imagine prompt: Renaissance oil painting, A king passing his crown to his son in a lavish throne room, With an atmosphere of grandeur and responsibility.
 
 3. /imagine prompt: Animated family movie style, A dad and son building a rocket ship in their backyard, Infused with a sense of adventure and bonding.
 
 4. /imagine prompt: Watercolor sketch, A father and son fishing on a quiet lake as the sun sets, With a serene and tranquil atmosphere.
 
 5. /imagine prompt: Comic book art, A superhero father with his sidekick son saving the day, With dynamic action and a sense of camaraderie.
 
 6. /imagine prompt: Traditional Japanese Ukiyo-e style, A samurai father and his young son practicing martial arts under a cherry blossom tree, With a sense of discipline and tradition.
 
 7. /imagine prompt: 90s sitcom illustration, A dad and son laughing together while attempting to fix a car in a garage, With a vibe of humor and togetherness.
 
 8. /imagine prompt: Modern abstract art, An interpretation of the emotional bond between a father and son, With swirling colors that convey deep connection and love.
 
 9. /imagine prompt: Children’s book illustration, A boy on his father’s shoulders, looking up at the stars, With a sense of wonder and shared curiosity.
 
 10. /imagine prompt: Sci-fi concept art, A father and son in futuristic gear exploring an alien planet, With a sense of discovery and shared adventure.

咱们能够简略地复制 / 粘贴这些提醒到 Midjourney。以下是其中的一些后果:

以上就是最简略 ChatGPT 的应用技巧,上面咱们还能够进行更高级的利用

裁减思路

因为 ChatGPT 应用了大量的语料库进行训练,所以咱们能够让它帮我咱们裁减思路,例如上面的提醒:

 keep theme, set prefix to "modern abstract art", 
 scene should contain metaphors (without explanations), 
 use suffix to implement lens and camera types. list 5 prompts

后果如下:

 /imagine prompt: Modern abstract art, An anchor and a soaring kite entwined, With a fisheye lens capturing the circular interplay of stability and freedom.
 
 /imagine prompt: Modern abstract art, A torch casting shadows of successive taller figures, With a tilt-shift lens emphasizing the central flame as the source of growth.
 
 /imagine prompt: Modern abstract art, Rivers of color weaving from a storm to a calm sea, With a wide-angle lens to envelop the viewer in the journey.
 
 /imagine prompt: Modern abstract art, A lighthouse beam morphing into a tapestry of music notes, With a telephoto lens bringing focus to the distant harmony.
 
 /imagine prompt: Modern abstract art, Clock gears turning into an unfolding map, With a macro lens capturing the intricacies of time and adventure's intertwinement.

后果如下:

优化 ChatGPT 输入

Midjourney 有一个叫 permutation 的性能,咱们在以前的文章中也做过介绍,所以能够应用上面办法,让 ChatGPT 生成相应格局的后果:

 This is the basic prompt anatomy for image generation with Midjourney:
 
 /imagine prompt: PREFIX, SCENE, SUFFIX
 
 PREFIX defines the image's medium and style
 SCENE defines the content
 SUFFIX modulates PREFIX and SCENE
 
 You can group variations in comma-separated lists like this:
 
 {PREFIX 1, PREFIX 2, PREFIX 3} or {SCENE 1, SCENE 2} or {SUFFIX 1, SUFFIX 2}
 
 Important: when using variations, each item in the list has to be a single word or a phrase without comma (otherwise we'd break the syntax)
 
 ###
 
 Create a prompt that illustrate the theme "father-son relationship". 
 Use 2 variations for prefix and 3 variations for suffix

这样 ChatGPT 就会返回上面的后果:

 /imagine prompt: {oil painting, watercolor}, a father teaching his son how to 
 ride a bike in a park, {with a touch of nostalgia, under a warm sunset, 
 captured in a moment of laughter}

这样,在将这个提醒复制到 Midjourney 时,会失去一下的后果:

也就是说,能够间接生成 6 张图片,以下是后果示例:

总结

ChatGPT 是通过大量语料库进行训练的大语言模型,而 Midjourney 则能够生成高质量的图像,通过联合 2 者的劣势实现更加综合和丰盛的内容创作。这种综合的内容生成形式可能通过文字和图像同时取得更多的信息和灵感。

https://avoid.overfit.cn/post/60d45f154b7943258f86f8bc7150e79b

作者:Tristan Wolff

正文完
 0