轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

DALL-E3 是目前所有AI绘画工具中,在生成一致性角色表现方面最出色的工具。下面我将介绍一些利用 DALL-E3 生成一致性角色的技巧。

一、基础操作,给图片创建种子编号

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

Prompt提示词 – 左

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a red t-shirt -XXXX

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着红色 T 恤 -XXXX

Prompt提示词 – 中

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt -XXXX

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤 -XXXX

Prompt提示词 – 右

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, and smiling -XXXX

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,面带微笑 -XXXX

重点:每个提示词都应以一个数字编号结尾,该数字编号从基数开始(例如,-0000),并针对每个变体递增。这就相当于给图片做了命名,或者理解为此编号充当图片的唯一标识符或种子。

二、让图片保持一致性并产生变化

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

这是编号为 0003 的图片,我将用它作为这篇文章的起始图片,在此图片的基础上做出一系列的变化,并且保持图片角色一致性!再次分享生成此图片的提示词:

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt -0003

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤 -0003

现在,我想让 Hope 微笑,只需要在提示词中增加 smiling 并且将图片编号递增即可。提示词如下:

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, smiling -0004

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,微笑着 -0004

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

怎么样?是不是比 Midjourney 强多了?可控性非常强,同时操作也很简单!这还没完,我们继续往下看。
现在,我想 Hope 有一根手指在她的嘴唇上。

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, smiling, and placing a finger on her lips -0005

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,微笑着,将手指放在嘴唇上 -0005

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

如果我想让 Hope 用手指在她的嘴唇上但不微笑,可以直接在对话框中向 DALL-E3 输入:take -0003 as base and now make Hope have a finger on her lips 我指的是-0003,因为在那个特定的图像中,她没有笑。所以,提示词是:

illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, placing a finger on her lips -0006

一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,将手指放在嘴唇上 -0006

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

三、总结归纳提示词

在前面的案例中,我们发现提示词的格式是有规律的。经过多次测试,我建议遵循这种格式,它效果最好:

[Base Prompt] + [additional details / variation information] -identifer+1

[基本提示] + [其他详细信息/变体信息] -识别编号+1

根据这个格式,我们可以给 ChatGPT 编写一个提示词,让它来遵循这个格式结构。

<—Prompt Start—>
Okay I have some instructions for you :
you can reframe the prompt in the following way:
Prompt Structure is : [Base Prompt] + [additional details / variation information] -identifer+1

Base Prompt: “Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt”

If i say make Hope do something base prompt will be above one which i provided .
If I say now make Hope do some thing , the base prompt will be the current state of Hope with the last identifier.

For any additional details or variations that don’t involve hair, append them under the “additional details / variation information” category and then add an identifier at the end.

Example 1: Changing Hair Style
If I ask for make hope hair flow, you can change the prompt to:
“Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair flowing, wearing a green t-shirt. -identifier+1”

Example 2: Coffee
If i ask to make Hope drink coffee, then you can change the prompt to:
“Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt and drinking coffee. -identifier+1”

This approach keeps the Base Prompt intact while allowing for specific variations as needed. The identifier at the end can help keep track of variations.
Are you clear with this ?
<— Prompt End—>

<—提示启动—>
好的,我有一些指示给你:
您可以通过以下方式重新构建提示:
提示结构为:[基本提示] + [其他详细信息/变体信息] -识别+1

基本提示:“一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤”

如果我说让希望做一些基本提示将高于我提供的提示。
如果我说现在让霍普做某件事,基本提示将是带有最后一个标识符的希望的当前状态。

对于不涉及头发的任何其他详细信息或变体,请将其附加到“其他详细信息/变体信息”类别下,然后在末尾添加标识符。

示例 1:更改发型
如果我要求让希望头发流动,您可以将提示更改为:
“一位名叫霍普的 30 岁美国女子的插图肖像,她的卷曲波浪卷发飘逸,穿着绿色 T 恤。-标识符+1”

示例 2:咖啡
如果我要求让希望喝咖啡,那么您可以将提示更改为:
“一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,喝咖啡。-标识符+1”

此方法保持基本提示完好无损,同时允许根据需要进行特定变化。末尾的标识符可以帮助跟踪变体。
你清楚吗?
<—提示结束—>

将上述提示词给到 ChatGPT 后,来自ChatGPT的回应:

<–Response Start —>
1. Yes, I understand the instructions clearly.
You’ve set a base prompt and provided a structure for any modifications.
2. I’ll use the current state of Hope as the base prompt if you use the word “now” and will modify it accordingly based on your specific requests, appending the variations under the “additional details / variation information” category. I’ll also increment the identifier for each new request.
3. Please provide your next instruction for Hope, and I’ll generate the image accordingly.
<— Response End —>

<–响应开始—>
1.是的,我清楚地理解了说明。
您已设置基本提示并为任何修改提供了结构。
2. 如果您使用“现在”一词,我将使用 Hope 的当前状态作为基本提示,并根据您的具体要求对其进行相应修改,在“其他详细信息/变体信息”类别下附加变体。我还将递增每个新请求的标识符。
3.请提供您对Hope的下一个说明,我将相应地生成图像。
<—响应结束—>

好的,现在我们来测试一下效果怎么样。
这是之前生成的 Hope 编号为0005的图片,我将用它作为现在的起始图片。

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

此时 Hope 的头是束起来的,我想让它散开。那么,可以直接在对话框中输入:now make Hope hair flow 此时的提示词将是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair flowing, wearing a green t-shirt, smiling, and placing a finger on her lips -0006

一位名叫霍普的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发飘逸,穿着绿色 T 恤,微笑着,将手指放在嘴唇上 -0006

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

可以在新生成的图片看到更新了头发细节,Hope 在微笑,因为那是她的在图片编号 0005 时的状态。
现在我们来试试给 Hope 加点东西,比如让她看书。直接在对话框中输入:Make Hope read book 这次我输入的提示中并没有用到 now 这个状态,看看结果会是怎样的。

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

提示词中没有使用 now 之后,DALL-E3 会将 0003 做为起始图片,此时的提示词将是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, reading a book -0007

一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,正在读书 -0007

目前的生成图片的编号是 0007 ,但是我想更新 0006 并让 Hope 喝咖啡,那么应该在对话框中输入:take -0006 as base and now make Hope drink coffee 请注意,0006 是散开的头发,如果要 Hope 喝咖啡的同时头发像 0003 一样束起来,那么应该在对话框中输入:take -0006 as base and now make Hope drink coffee but in bun hairstyle

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

此时,左图的提示词则是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair flowing, wearing a green t-shirt, smiling, placing a finger on her lips, and drinking coffee -0008

一位名叫 Hope 的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发飘逸,穿着绿色 T 恤,微笑着,将手指放在嘴唇上,喝咖啡 -0008

右图的提示词是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a green t-shirt, smiling, placing a finger on her lips, and drinking coffee -0009

一位名叫 Hope 的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻,穿着绿色 T 恤,微笑着,将手指放在嘴唇上,喝咖啡 -0009

四、DALL-E3 角色一致性提示词技巧进阶

前面,我们利用提示词解决了生成角色一致性的问题,并且总结归纳了提示词结构,还在此基础上给生成图片添加了不同的无素。现在我们要生成 Hope 从事各种户外或室内活动的图片,在开始生成图片之前需要让 ChatGPT 考虑一些重点问题,给它更新一下提示词结构吧。

<—Prompt Start—>
Now, we’re going to work on the activity scene.
We will set the base prompt as: ‘Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun.’
Is that okay? In this case, you will have the freedom to update clothes and other details under the ‘additional elements’ section.
Are you ready now?
<— Prompt End—>

<—提示启动—>
现在,我们将在活动场景上工作。
我们将基本提示设置为:“一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻。
可以吗?在这种情况下,您可以在“附加元素”部分下自由更新衣服和其他详细信息。
你现在准备好了吗?
<—提示结束—>

ChatGPT 的回应:

<–Response Start —>
Yes, that sounds like a great approach. Setting the base prompt as:
“Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun.”
gives flexibility to adjust attire and other elements based on the activity or scene without altering the primary character features. I’ll ensure to make modifications only within the “additional details / variation information” category based on the activity you specify.
Thank you for the clarification, and I’m ready to proceed with this approach. Please provide the next activity or scene you’d like for Hope.
<— Response End —>

<–响应开始—>
是的,这听起来是一个很好的方法。将基本提示设置为:
“一位名叫霍普的 30 岁美国女性的插图肖像,她的卷曲波浪卷发梳成发髻。”
可以根据活动或场景灵活地调整服装和其他元素,而无需更改主要角色特征。我将确保仅根据您指定的活动在“其他详细信息/变体信息”类别中进行修改。
感谢您的澄清,我已准备好继续采用这种方法。请为希望提供您想要的下一个活动或场景。
<—响应结束—>

进行此调整是因为 Hope 穿着绿色 T 恤,而实际需求是不会穿着一成不变的服装的。与此同时,还让 ChatGPT 修改起始图片Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun 现在,每一张新生成的图片都将遵循这个提示词作为起始图片。现在我想试试让 Hope 去野外登山,在对话框中输入:Hope doing hiking ,此图片的提示词将会是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, hiking in the mountains with appropriate attire and gear -0010

一位名叫 Hope 的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发梳成发髻,穿着合适的服装和装备在山上徒步旅行 -0010

可以看到 ChatGPT 自动添加了细节,它使用了 mountain appropriate attire 作为细节描述词。

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

既然 ChatGPT 帮我们丰富了图片的细节,那能不能指定 Hope 服装的颜色呢?我们在对话框中输入命令来试一试:Hope doing hiking but I want her to wear white jacket and red backpack 此时提示词将会是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, wearing a white jacket and carrying a red backpack, hiking in the mountains -0011

一位名叫霍普的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发梳成发髻,穿着白色夹克,背着红色背包,在山上徒步旅行

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

效果不错,DALL-E3 完全能理解我们的要求。现在,我来试试变换场景,让 Hope 在冬天堆雪人。对话框中输入命令:Hope in a winter setting, making a snowman 此时提示词将会是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, dressed in winter attire, in a snowy landscape making a snowman -0012

一位名叫霍普的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发梳成发髻,穿着冬季服装,在雪景中堆雪人 -0012

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

最后测试一下 DALL-E3 生成带指定文字的图片,Hope 在图书馆里,穿着一件写着“Hope”的黑色T恤,输入命令:Hope is in the library, wearing a black t-shirt with 'HOPE' written on it 提示词将会是:

Illustration portrait of a 30-year-old American woman named Hope with her curly wavy hair styled in a bun, seated comfortably in a library setting, wearing a black t-shirt with the word ‘HOPE’ written on it, engrossed in reading a book -0013

一位名叫霍普的 30 岁美国妇女的插图肖像,她的卷曲波浪卷发梳成发髻,舒适地坐在图书馆环境中,穿着一件写有“Hope”字样的黑色 T 恤,全神贯注地阅读一本书 -0013

轻松驾驭生成图片一致性,DALL-E3最佳解决方案!

未完待续…

相关文章