/adt/ - Anime Diffusion Thread #130

Previous: >>109538028

上一篇：>>109538028

NEW:

https://huggingface.co/Comfy-Org/MiniMax-H3

Minimax H3 (video model) is out

Minimax H3（视频模型）已推出

https://huggingface.co/circlestone-labs/Anima

Anima1.0 is out

Anima1.0 发布了

>UIs

ComfyUI:https://github.com/comfyanonymous/ComfyUI

SwarmUI:https://github.com/mcmonkeyprojects/SwarmUI

KritaAI: https://github.com/Acly/krita-ai-diffusion

Forge Classic/Neo: https://github.com/Haoming02/sd-webui-forge-classic

>How to generate anime:

>如何生成动漫风格图片：

image : https://rentry.org/comfyui_guide_1girl

video: https://rentry.org/wan22ldgguide

>Style Explorers

>风格探索者

Anima: https://anima.mooshieblob.com/ | https://animadex.net/

Illustrious: https://tagexplorer.github.io

>Checkpoints, LoRAs, Upscalers, & Workflows

>检查点（Checkpoints）、LoRA、放大器和工作流

https://civitai.com

https://civitaiarchive.com

https://tensor.art

https://openmodeldb.info

https://www.seaart.ai

https://www.liblib.art/

>Resources:

Share Metadata: https://litterbox.catbox.moe

Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-beta-one

Online metadata viewer SD/NovelAI: https://spell.novelai.dev

Catbox/Metadata Userscript: https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fb

Output cleanup: https://rentry.org/RemovingDiffusionGunk , https://www.mediafire.com/file/vipr23exc5htmnt (batch processing python script)

Inpainting Guide from an Anon: https://files.catbox.moe/fbzsxb.jpg >>106520607

>/adt/ logo:

>/adt/ 标志：

https://files.catbox.moe/pakfgt.png

>Beware inorganic activity

>警惕无机活动

https://rentry.org/animanon

图片: https://i.4cdn.org/g/1786813118119307.jpg

https://www.anistudio.com/

First for (tr)ani is literal human garbage

(tr)ani 首先是字面上的人类垃圾

>>109564413

The website is top notch

该网站是一流的

(无正文)

图片: https://i.4cdn.org/g/1786816329914342.png

thanks for bakering

感谢烘焙（训练）

(无正文)

图片: https://i.4cdn.org/g/1786821898029495.jpg

Last thread had some nice oppai. Let's keep it going anons.

最后一个线程有一些不错的oppai。让我们继续保持下去吧。

>>109565796

you like boobs?

你喜欢胸部吗？

图片: https://i.4cdn.org/g/1786831312222733.jpg

get on all fours anon

立刻四肢着地

图片: https://i.4cdn.org/g/1786835808143238.png

(无正文)

图片: https://i.4cdn.org/g/1786836956982296.png

>>109564235

How do I reach this level of prompting?

我如何才能达到这种提示级别？

>https://www.pixiv.net/en/users/120866362

Just give me general tips for a retard that is using plain Anima aesthetics basic model. I would like to create comics & illus that look as good. If I wanna create comic panels like pic rel, I'm obviously supposed to prompt separately and then edit everything in CSP, right? Don't bully me

只需给我一些关于使用普通阿尼玛美学基本模型的弱智的一般提示即可。我想创作看起来同样好的漫画和插画。如果我想创建像 pic rel 这样的漫画面板，我显然应该单独提示，然后在 CSP 中编辑所有内容，对吗？别欺负我

图片: https://i.4cdn.org/g/1786839465909108.jpg

(无正文)

图片: https://i.4cdn.org/g/1786839570294363.jpg

>>109567493

you answered your own questions

你回答了你自己的问题

one-shot comic gens are not a thing

一次性漫画一族不是一回事

>>109567493

>How do I reach this level of prompting?

>我如何才能达到这种提示级别？

You don't. This is editing. And each page is several gens.

你不知道。这是编辑。而且每一页都是好几代。

>>109567561

Just wanted to make sure

只是想确定一下

>>109567565

>You don't. This is editing

>你不知道。这是编辑

Can you elaborate more? I mean besides the comic formatting.

您能详细说明一下吗？我的意思是除了漫画格式之外。

The style and the gens just look super good

风格和风格看起来都超级好

How do I use krea2 with forge? What TE things do I need?

如何将 krea2 与 forge 一起使用？我需要哪些 TE 物品？

图片: https://i.4cdn.org/g/1786855415994898.jpg

reminder cache is king, expensive compute tasks for loading models can be done once by saving to cache files and forever load it in <1s instead of many seconds/minutes (prompt conditioner, tokenizer, gguf meta data, clip, vae encoder latent)

提醒缓存是王道，加载模型的昂贵计算任务可以通过保存到缓存文件一次完成，并在 <1 秒内永远加载它，而不是许多秒/分钟（提示调节器、分词器、gguf 元数据、剪辑、vae 编码器潜在）

>>109568992

and also for wan video generation, for those using 4step lora for fast generation, TeaCache works best if you skip the first and the last step and only run in the 2nd and 3rd step. not 1/4. With 0.25, the inference speed for total cuts down by almost half with minimal quality difference. If you do TeaCache for all steps, you get fucked up quality outputs. So smarter to only use TeaCache for the middle steps only.

对于 wan 视频生成，对于那些使用 4step lora 进行快速生成的人来说，如果跳过第一步和最后一步，只运行第二步和第三步，TeaCache 的效果最好。不是 1/4。使用 0.25 时，总推理速度几乎减少一半，质量差异最小。如果所有步骤都使用 TeaCache，那么输出的质量就会很糟糕。因此更明智的做法是仅将 TeaCache 用于中间步骤。

REAL THREAD IS HERE

>>109568950

>>109568950

>>109568950

>>109568950

>>109569076

Where

>>109568967

The same setup as Comfy, you have to load the VAE and TE in the same tab

与 Comfy 相同的设置，您必须在同一选项卡中加载 VAE 和 TE

>>109568967

Why not open up an AI desktop workspace and ask it to install this shit on your computer? Thats the best thing AI can do for tech illterates

为什么不打开一个人工智能桌面工作区并要求它在你的计算机上安装这个东西呢？这是人工智能能为科技盲人做的最好的事情

>>109568967

got it, I used these. It's really slow for me though it's so over....

明白了，我用过这些。虽然已经结束了，但对我来说真的很慢......

图片: https://i.4cdn.org/g/1786861505539385.png

>>109569450

AI difusion is about make friends and make finetunes and loras for them to larp

AI扩散是为了交朋友并制作微调和洛拉让他们朗读

>>109569480

Really? For me Krea turbo at 10 steps is faster than Anima.

真的吗？对我来说，Krea 涡轮增压 10 步比 Anima 快。

>>109567493

You'll be able to do this with NovelAI v5 once it releases.

一旦 NovelAI v5 发布，您就可以做到这一点。

>>109569503

Huh really? It's slower at 8 steps than anima at like 30+. I got it fixed somewhat but it's still pretty damn slow.

嗯，真的吗？ 8 步时它比 30 步以上的阿尼玛慢。我已经把它修复了一些，但它仍然非常慢。

>>109569554

Are you using confidence gating early exit, teacache, turbo lora?

您是否使用信任门提前退出、teacache、turbo lora？

>>109569561

nope what's that

不，那是什么

Official CirclestoneLabs™ Anima genning competition is on!.

官方 CirclestoneLabs™ 动画生成竞赛开始了！

Will /adt/'s pedigree genners show up?

/adt/的血统家族会出现吗？

https://kusart.com/zh-CN/events/anima

图片: https://i.4cdn.org/g/1786863132270960.jpg

(无正文)

图片: https://i.4cdn.org/g/1786863588059683.png

>>109569634

Oh cool anon, it's been a while since I visited /ldg/, but I'm not moving from here, you know, I'm more of an /adt/ kind of girl teehee~

哦，酷啊，我已经有一段时间没有访问/ldg/了，但我不会离开这里，你知道，我更像是一个/adt/类型的女孩teehee~

>>109567112

Pretty sure that's Diving Suit Anima.

很确定那是潜水服阿尼玛。

>>109569577

Repost thread tho

重新发布主题

(无正文)

图片: https://i.4cdn.org/g/1786872151894500.jpg

>>109568967

>https://civitai.red/models/2862187/final-fantasy-tactics-akihiko-yoshida-style-krea2-lora?modelVersionId=3233281

lol now it makes sense why you wanted Krea 2

哈哈，现在你明白为什么你想要 Krea 2 了

>>109570584

god that looks so fucking bad

天哪，看起来太糟糕了

>>109570636

Why

>>109570641

do you not have eyes?

你没有眼睛吗？

>>109569527

So in like 5 years.

所以大约5年之内。

Stupid cunts just sit on it.

蠢货们就坐在上面吧。

>>109570648

I have eyes

我有眼睛

>>109570669

then why the fuck are you asking why something that fried looks bad. Are you retarded? Do you have no aesthetic sense? Do you not know your left from your right?

那你他妈的为什么要问为什么油炸的东西看起来很糟糕。你是弱智吗？你没有审美意识吗？你不知道你的左手和右手吗？

>>109570681

I like it, it's a good story, but you can't explain why you don't like it.

我喜欢它，这是一个好故事，但你无法解释为什么你不喜欢它。

Why you don't like the lora, anon?

为什么你不喜欢劳拉呢？

>>109570681

then why the fuck are you asking why something that fried looks bad.

那你他妈的为什么要问为什么油炸的东西看起来很糟糕。

>Are you retarded?

>你是弱智吗？

Ad hominem slop

人身攻击

>Do you have no aesthetic sense?

>你没有审美吗？

Ad hominem slop

人身攻击

>Do you not know your left from your right?

>你不知道你的左边和你的右边吗？

Ad hominem slop

人身攻击

>konan is now trying to expand his grift to a shitty niche thread on /g/

>konan 现在正试图将他的欺骗扩大到 /g/ 上的一个糟糕的利基线程

absolutely grim

绝对严峻

>>109570744

bro actually took a look at the lora and couldn't keep his post up in good faith.

兄弟实际上看了一眼lora，并不能善意地保留他的帖子。

>>109564235

No idea who the baker is and don't really care, but this needs to come out of the OP:

不知道面包师是谁，也不在乎，但这需要从OP中得出：

>video: https://rentry.org/wan22ldgguide

WAN is dead and outdated, and it's got 'ldg' baked into the link, so anyone searching for ldg ends up finding us instead, not a fan of that, we're not affiliated with them in any way. We're /adt/, just another island on /g/ like /aicg/ or /de3/. We don't need "thread neighbours" nonsense or some shared avatarfag history with other threads.

WAN 已经死了并且过时了，并且它的链接中包含了“ldg”，因此任何搜索 ldg 的人最终都会找到我们，而不是我们的粉丝，我们与他们没有任何关系。我们是 /adt/，只是 /g/ 上的另一个岛屿，就像 /aicg/ 或 /de3/ 一样。我们不需要“线程邻居”废话或与其他线程共享一些化身历史。

Also kill this dumb link while we're at it:

当我们这样做时，还要杀死这个愚蠢的链接：

>Beware inorganic activity

>警惕无机活动

https://rentry.org/animanon

>>109570779

Good point, I'll replace the WAN link for a Minimax H3 link.

好点，我将用 Minimax H3 链接替换 ​​WAN 链接。

>stealth ani-posting

> 秘密发布

i nooticed

我注意到

Wanted to make some cool anime images, so I went and downloaded like 10 loras off civitai that all looked amazing in the sample pics. Turns out like 90% of them look like dogshit once you use them. >:(

想要制作一些很酷的动漫图像，所以我去 civitai 下载了大约 10 个洛拉，它们在示例图片中看起来都很棒。事实证明，一旦你使用它们，其中 90% 看起来都像狗屎。 >:(

图片: https://i.4cdn.org/g/1786884588837278.jpg

>>109570584

I see new fft lora I get new model

我看到新的 fft 洛拉 我得到了新模型

although I did skip z-image and flux.

虽然我确实跳过了 z 图像和通量。

图片: https://i.4cdn.org/g/1786884707279749.jpg

>>109571030

not even sure why they bother being like that tho, cuz whether you use your lora a lot or not, nobody's earning steady money off this anyway

甚至不知道他们为什么要那样麻烦，因为无论你是否经常使用你的劳拉，反正没人能靠这个赚到稳定的钱

>>109571040

What's your take on Krea 2? At high frequencies I feel Anima still holds onto aesthetic and style better, whereas Krea is more consistent in that same range, though that's probably just down to the parameters.

您对《克里亚2》有何看法？在高频下，我觉得 Anima 仍然能更好地保持美学和风格，而 Krea 在同一范围内更加一致，尽管这可能只是取决于参数。

>>109570779

KIll yourself Julien

自杀朱利安

(无正文)

图片: https://i.4cdn.org/g/1786885314199358.png

>>109571086

I see some potential but I don't know what I'm doing yet and I can't generate large breasts. I'm not sure what you meant by frequencies or parameters.

我看到了一些潜力，但我不知道我在做什么，而且我无法产生大乳房。我不确定你所说的频率或参数是什么意思。

图片: https://i.4cdn.org/g/1786885615331731.jpg

>>109570740

>shitty niche thread on /g/

> /g/ 上的垃圾利基线程

Oh yeah? You think you are so superior with your "quality" thread? You probably jack off to your ComfyAnonymous dakimakura every night while crying about how misunderstood you are, pathetic /ldg/ troon.

哦，是吗？你以为你的“质量”线程就那么优秀吗？你可能每天晚上都会在 ComfyAnonymous dakimakura 上自慰，同时哭泣着自己被误解了，可悲的/ldg/troon。

>>109570822

>>109571103

I support Ani because I love arguing with braindead morons who hate him, they are just as much of a bottom feeding retard as they are.

我支持阿尼，因为我喜欢和那些讨厌他的脑残白痴争论，他们和他们一样都是底层喂养的白痴。

.

>>109570822

>>109571103

oh wow I thought you guys were schizoing out until now but you were right

哦哇，直到现在我还以为你们精神分裂，但你们是对的

I apologize

我道歉

>>109571149

the fuck are you even on about. konan is a known grifter who has stolen shit from people on /h/ in the past. That statement has nothing to do with your retarded thread war shit.

你他妈的还在想什么。 konan 是一个著名的骗子，他过去在 /h/ 上偷过别人的东西。这种说法与你的弱智线程战争无关。

>>109571149

Again, you should be making peace with your god while you still can, Julien

再说一次，你应该趁你还可以的时候与你的上帝和解，朱利安