CONSISTENCY | Midjourney Office Hours Recap Jan 31st 2024 | Midjourney News

Summary

在这段1月31日的中途办公时间回顾中，主要讨论了即将发布的官方Beta 6版本，包括改进的图像质量和审美变化。还提到了可能加入的新功能，比如结合图像提示和风格调整器的“参考图像”功能，旨在提升图像生成的一致性。此外，它们还将开放更多对Alpha创建网站的访问，并计划改进主网站的探索页面。团队正努力多样化顶级图像的展示，避免将其变成带有个性化图像建议的Instagram feed。另外，他们正在改进描述功能，以更准确地识别重要的语言模式，并计划在未来几周内测试并推出。此外，还有关于改善网站设计和功能的讨论，以及收集的反馈汇总，显示字符一致性是用户最关心的问题。

Takeaways

• 😀 正在准备发布6.0正式内测版,图像质量略有改进,美感也有调整
• 😊 更灵活的参考图片功能即将推出,可以从图片中提取元素应用到生成图像
• 😎 一致性功能正在开发中,首先实现样式一致性,人物一致性后面再做
• 😮 新的描述界面即将面向指导员和版主测试,提高提示的准确性
• 😀 pan zoom和绘画功能可能下周推出
• 😀 面部生成一致性是重点开发内容
• 😊 网站设计和功能也在进行改进
• 😕 生成人物时避免艺术裸体存在分歧
• 😀 提升远处人脸效果的方法已经探索出来
• 😊 文本渲染质量将得到提升

Q & A

本周Midjourney可能会推出哪项新功能?

可能会推出一项新功能,将参考图片和风格调整相结合,被称为参考图片。它会从参考图片中获取某些特定元素,并将其应用到后续生成的图像中。这是 Midjourney 一致性功能的开始,会让图像在不同生成中在视觉上保持一致。

David对风格调整这个名称有何看法?

David认为这个名称可能有问题,因为它与模型本身关系更大,而不是真正的风格。所以名称可能会有所改变。

新版描述功能有何进步?

新版描述功能更加准确,但不总是完美。它可能会通过识别更重要的语言模式来帮助人们更好地进行提示。

人物一致性功能什么时候会推出?

在推出所有这些新功能后,他们会着重关注人物一致性,特别是面部。这是下一个重点。

提升画质距离加入多数用户关注前十?

改善远距离人脸画质进入了前十。他们对如何修复这一问题有了想法,可能会在7版本中实现。

艺术裸体在用户反馈中情况如何?

许多人要求艺术裸体,但也有很多人不同意。所以这一功能并不确定。

Midjourney下一步工作重点是什么?

他们现在关注基础功能,想吸引更多新用户进行提示。所以不太关注高级用户的定制功能。

背景透明度是否会实现?

他们知道如何实现,但不打算在7版本中关注这一功能。现在关注7版本的其他方面。

画作是否可以隐藏和批量删除?

未来可能可以隐藏画作,但可能无法批量删除。

控制网目前进展如何?

本月他们正在训练一些东西,但不知道是否会有效。

Outlines

00:00

📢 中途办公时间1月31日回顾

本视频回顾了1月31日中途办公时间的主要内容，包括即将发布的6版本官方Beta版，该版本在图像质量和审美上有所改进，并可能加入新功能。此外，将开放更多用户访问alpha创建网站，新的探索页面即将上线主网站，同时致力于多样化顶部图片以避免形成类似Instagram的个性化图片推荐。重大功能更新可能包括结合图像提示和样式调整器的参考图像功能，以及对一致性特性的初步实现，尤其是在风格一致性方面。对于风格调整的未来，可能会重新考虑其命名和功能定位。新的描述工具正在改善中，计划先与指南和管理员测试，预计下周向所有用户推出。Nii模型将增加平移、缩放和修复功能。在发布这些新功能后，将专注于改进角色一致性，特别是面部表现。网站设计和功能也将得到改进。收集到的反馈包括对角色一致性、提示理解、人类表现改进等方面的要求，以及对艺术裸体、远距离人脸、内部填充、图片隐藏和提升放大功能的建议。版本7将包含一些这些改进，而有些功能如透明背景和更复杂的上采样可能不会立即实现。风格参考和颜色控制将得到改善，但网站的组织功能和用户配置文件跟踪可能不是当前的重点。视频还提到了艺术作品的制作提示，并鼓励观众在Instagram上关注以获取更多图片。

Keywords

💡mid-journey

mid-journey是一个AI图像生成系统,用户可以用文字描述一个想法,mid-journey就会根据描述生成相关的图像。视频中提到了mid-journey最近在开发的新功能,以及未来规划等信息。

💡图像质量

视频提到mid-journey正在改进图像质量,新版Beta 6会比之前版本有略微提升的图像质量和美学改变。这显示mid-journey正在持续优化其图像生成算法。

💡创建网站

mid-journey正在开发一个图像创作网站,允许更多用户访问。这有助于mid-journey收集更多用户反馈,改进算法。视频中表示这个新网站的“探索”页面即将上线。

💡个性化

mid-journey希望避免主页图片流成为类似instagram的风格,而是提供个性化的图像建议。这显示mid-journey注重用户体验的个性化和多样性。

💡参考图像

视频提到mid-journey可能会推出一个新功能,允许用户上传参考图像,系统会分析图像的特定风格或元素,应用到后续生成的图像中。这是mid-journey正在研发的图像风格一致性功能的第一步。

💡风格调整

视频中提到风格调整功能可能会改名,因为它更关注模型而不是风格。风格参考图像功能可以被看作是更灵活版本的风格调整。这显示mid-journey正在迭代这个功能。

💡人物一致性

视频表示人物一致性是用户高频请求的功能之一,但相比风格一致性更难实现。mid-journey在推出图像参考功能后,会聚焦在改进人物一致性,特别是人脸部分。

💡用户反馈

视频列出了mid-journey收集到的用户功能请求排名前几位,包括人物一致性,提示理解等。这显示mid-journey高度重视用户反馈,以此来确定产品路线图优先级。

💡版本更新

视频中多次提及版本号,说明mid-journey正在快速迭代产品功能。新版本通常会带来图像质量提升、新功能等。这显示mid-journey的快速发展步伐。

💡艺术创作

视频中用户请求支持艺术创作,但也有用户反对。这显示mid-journey擅长图像生成,但在处理具有争议题材时,需要谨慎并关注各方反馈。

Highlights

正式版6预计本周发布,图像质量略有提升,美学变化

更多访问alpha创作网站,新的探索页面将在主网站推出

本周可能推出参考图片功能,应用图片的特定样式到生成图片

一致样式功能即将到来,使生成图片视觉上保持一致

新描述功能测试良好,可能下周面向所有用户

面部一致性将成为新版本重点

网站设计和功能也将得到改进

人物一致性是用户反馈中的第一要求

理解提示内容是第二要求,正在针对第7版本改进

正致力于皮肤纹理的改进,可能在第7版本之前实现

远距离脸部效果改进,可能在第7版本体现

上采样功能很难不变复杂的情况下显著提升

不太可能支持大规模删除图片

第7版本将专注基础功能,而非大规模组织功能

更看重新用户,而非功能偏好高级用户

Transcripts

hey everyone here's a recap of the mid

Journey office hours from January 31st

stick around until the end if you want

to hear a recap of all the feedback

they've collected in the past week okay

the main things they're working on are a

version 6 official Beta release with

slightly improved image quality and

aesthetic changes there might be another

feature or two coming but I'll get to

that later as well they will also be

allowing more access to the alpha

creation website the new explore page

will be put on the main website soon and

they're still trying to diversify what

people see in the top images they want

to avoid making it an Instagram feed

with personalized image suggestions the

one major feature we might see this week

and maybe later today as in the night

before you're hearing this it's going to

be a cross between image prompting and a

style tuner I think they're going to

call it reference images where it will

take a specific thing from the image and

apply it to Future generations and this

is the beginning of the consistency

features David said consistent styles

were easier so that's what's going to

come out first versus character

consistency which we might see later

consistent Styles will try to make

things visually consistent across

Generations it might not be perfect it's

going to be kind of like style tuning

but a little more flexible because you

can provide any single image as far as

the future of style tuning he said there

may be a problem with the name because

it wasn't really about style it was more

about the model so we'll see how that

changes he also said that the new

described is starting to look and work

better they're going to test it out with

the guides and mods this week maybe ship

it to everyone next week it's more

accurate but not always perfect and it

might help people prompt better by

identifying more important language

patterns the nii model will get pan zoom

and in painting maybe next week then

they will try to focus on character

consistency after all of these new

releases with an emphasis on faces also

we can expect some improvement to the

website design and function okay to the

collected feedback he showed us the

visualization and the list of all the

requested features character consistency

is number one so that is what they're

working on prompt understanding is

number two David said they think they

know the bottleneck and are working to

correct that for version 7 number three

was better humans and they're already

working on improvements to skin texture

we might see that before version 7 lots

of people were asking for artistic

nudity but lots of people disagreed with

that in the feedback so take that for

what you will people requested better

faces at a distance they have an idea on

how to fix that and it might make it to

version 7 improved in painting was next

and he said it's kind of hard to

understand what people actually want

with that so they'll have to ask more

questions David said we will be able to

hide images in the future but probably

won't ever be able to delete them on

mass better UPS scales will be coming he

said it's hard to make an upscaler way

better without making it really

complicated text rendering will improve

as far as control net goes they are

training some stuff this month but they

have no idea if it will work transparent

backgrounds were requested they know how

to do it but I'm saying he didn't really

sound convinced or inspired to do it

they want to focus on version 7 version

5 style tuning won't come back but style

reference can be used with those older

pictures and it might kind of work hex

code color control that could work and

they have some ideas on how to implement

it more organization features maybe not

they are focused on the basics for now

following user profiles they do not want

to copy Instagram bigger grids maybe in

version 7 better seamless tiling he said

he isn't convinced it's really any

different in version 6 maybe it's just a

prompt issue exporting our prompt

history David said they're focused on

newcomers more than power users they

need more people to prompt more so maybe

eventually if you like the art in this

video here is the prompt it's the

subject in the style of large scale

painting Paul corfield conceptual

digitalism contemporary landscape John

MCN monochromatic masterpieces Patrick

Brown I used styze 40 100 400 and style

raw I used chaos 6 and some of these

were made inside of nii version 6 which

was just launched and I don't think we

talked about it in this office hours

recap go play around with that it's a

lot of fun I also got this prompt from

gayg over on the mid Journey website so

thanks a bunch if you want a better look

at some of my pictures you can follow me

over on Instagram @ future techpilot

there's your recap of the mid-journey

office hours I hope you're doing well

take care and I'll see you next time