She's an Instagram model from Barcelona who's been going viral, but she's also just like totally chill and down-to-earth.

她是一名來自巴塞羅那的 Instagram 模特，在網上走紅，但她也是一個完全冷靜、腳踏實地的人。

She's into fitness and video games, and yeah, if you're wondering, she does have additional photos on her subscription tier, which brings in about 10 grand per month.

她喜歡健身和電子遊戲，是的，如果你想知道，她的訂閱層級確實有額外的照片，每月能帶來約 1 萬美元的收入。

Before you sign up though, there's one thing I need to disclose, and it really shouldn't matter unless you're some kind of bigot technophobic Nazi, but she's not a biological female.

不過，在你註冊之前，有一件事我需要透露一下，除非你是某種偏執的技術恐懼症納粹分子，否則這真的不重要，但她不是生理上的女性。

She is a woman, but entirely artificial.

她是一個女人，但完全是人造的。

It is November 29th, 2023, and you're watching The Code Report.

現在是 2023 年 11 月 29 日，您正在收看《密碼報告》。

By the end of this video, you'll know how to build your own army of AI influencers entirely for free using open source generative image models like Stable Diffusion XL and checkpoints like Juggernaut that can produce highly realistic images.

在本視頻結束時，您將知道如何使用穩定擴散 XL 等開源生成影像模型和 Juggernaut 等可生成高度逼真影像的檢查點，完全免費地組建自己的人工智能影響者大軍。

The first generative adversarial networks appeared about 10 years ago, and at first could only produce tiny images that you could barely make out, but now anyone can produce high-resolution realistic images good enough to trick sad lonely men into buying the NSFW content.

第一批生成式對抗網絡大約出現在 10 年前，起初只能生成幾乎看不清的微小影像，但現在任何人都能生成高分辨率的逼真影像，足以欺騙悲傷的孤獨男人購買非色情內容。

It's extremely sad and depressing that the agency behind this thing is making a month of No Nut November.

這背後的機構將 "無堅果 11 月 "作為一個月度活動，這讓人感到非常悲哀和沮喪。

Content like this is rotting the brains of billions of men around the world.

這樣的內容正在腐蝕全世界數十億人的大腦。

Although they are doing a 50% off sale right now, so it does look like a pretty good value actually.

雖然他們現在正在打五折促銷，但看起來確實很划算。

Luckily, 95% of this channel is comprised of high-value alpha males, which means we don't need to consume content like that.

幸運的是，這個頻道中 95% 的人都是高價值的阿爾法男性，這意味著我們不需要消費這樣的內容。

Instead, we reverse engineer it so we can exploit it for a financial gain.

相反，我們對其進行逆向工程，以便利用它來獲取經濟利益。

There are many tools out there that can help you become an AI PIMP, like you might be familiar with Midjourney and Dolly from OpenAI, but these are paid, closed-source products that also contain a bunch of unwanted safety layers.

市面上有很多工具可以幫助你成為人工智能 PIMP，比如你可能熟悉的 OpenAI 的 Midjourney 和 Dolly，但這些都是付費的閉源產品，還包含一堆不必要的安全層。

Luckily, there's a great open-source ecosystem around generative AI.

幸運的是，圍繞著生成式人工智能有一個很好的開源生態系統。

The most well-known base model is Stable Diffusion XL, which was released in late July 2023.

最著名的基礎型號是穩定擴散 XL，於 2023 年 7 月底發佈。

Gigantic models like this take an absurd amount of computing power, but luckily it's possible to fine-tune them by creating what's called a checkpoint using some additional specialized training data.

像這樣的巨型模型需要耗費大量的計算能力，但幸運的是，通過使用一些額外的專業訓練數據創建所謂的檢查點，可以對其進行微調。

Now you don't have to worry about this part, because people have already created a bunch of different checkpoints which you can find on websites like Civit.ai, some of which are highly optimized for photorealism.

現在您不必擔心這部分了，因為人們已經創建了許多不同的檢查點，您可以在 Civit.ai 等網站上找到，其中一些檢查點已經針對逼真度進行了高度優化。

Now the next part of the puzzle is finding a UI where we can actually work with these models without having to write any code.

現在，謎題的下一部分是找到一個用戶界面，讓我們無需編寫任何代碼就能使用這些模型。

The most well-known option is the Stable Diffusion Web UI.

最著名的選項是 Stable Diffusion Web UI。

It's an incredibly powerful tool, but might be a little overwhelming for a beginner.

它是一個非常強大的工具，但對於初學者來說可能會有點力不從心。

There's just a lot of different buttons to push.

有很多不同的按鈕可以按。

Another cool option is Comfy UI, which has a drag and drop editor that works kind of like Blender or Unreal Engine.

另一個很酷的選擇是 Comfy UI，它有一個拖放編輯器，工作原理有點像 Blender 或虛幻引擎。

But the one I'm using in this video is Focus, or Fucus, and I'll explain why I like it in just a minute.

但我在本視頻中使用的是 Focus 或 Fucus，稍後我會解釋我為什麼喜歡它。

You'll notice all these AI UIs look very similar, and that's because they're based on another open-source project called And an interesting side note is that Gradio builds the front end with Svelte.

你會發現所有這些人工智能 UI 看起來都非常相似，這是因為它們都基於另一個開源項目，名為 Gradio。

Pretty cool, but let's get started.

很酷，但讓我們開始吧。

Check out the install instructions, but basically all you do is clone the repo, create a virtual environment for Python, and install the packages.

請查看安裝說明，基本上你要做的就是克隆軟件倉庫，為 Python 創建一個虛擬環境，然後安裝套裝軟體。

Once that's done, you run this Python script, which will then automatically download the default models in the background.

完成後，運行這個 Python 腳本，它就會在後臺自動下載默認模型。

These files are multiple gigabytes large, so it may take a while.

這些文件有多個千兆字節大，是以可能需要一段時間。

The base model is Juggernaut XL, which is based on Stable Diffusion and works pretty well for realistic images.

基本模型是 Juggernaut XL，它基於穩定擴散技術，在處理逼真影像時效果相當不錯。

You now have the power at your fingertips to prompt out any image you can imagine.

現在，您的指尖就能發出您能想象到的任何影像。

The thing I like about Focus is that in many ways it feels similar to Midjourney, but it's free and the UI is way more intuitive.

我喜歡 Focus 的一點是，它在很多方面都與 Midjourney 相似，但它是免費的，用戶界面也更直觀。

I'm running it with a modest Nvidia 3070, and it usually takes about 45 seconds to generate two quality images.

我使用的是性能一般的 Nvidia 3070，生成兩張高質量影像通常需要大約 45 秒。

What's really cool though is that if you click on the Advanced tab, you'll get options for performance, aspect ratio, and the number of images, but there's also a tab for style, and we can mix and match these different styles to get far different results, like this retro video game style, or maybe an anime style.

最酷的是，如果點擊 "高級 "選項卡，你會看到性能、寬高比和圖片數量等選項，而且還有一個風格選項卡，我們可以混合搭配這些不同的風格，以獲得截然不同的效果，比如復古的視頻遊戲風格，或者動漫風格。

But in order to generate our AI Influencer, I'm going to go with the default settings along with Focus Photograph.

不過，為了生成我們的人工智能影響者，我將使用默認設置和 "焦點照片"。

Create a prompt that's highly specific, and add some imperfections like rough skin, no makeup, etc. to make it more realistic.

創建一個高度具體的提示，並添加一些瑕疵，如皮膚粗糙、沒有化妝等，使其更加真實。

And now our Influencer is born.

現在，我們的 "影響者 "誕生了。

Now let's go ahead and save this as our base image, then create a new prompt and check the box for Input Image.

現在，讓我們將其保存為基礎影像，然後創建一個新的提示，選中輸入影像複選框。

Drop the base image here to blend multiple images and text together.

將基礎影像放在這裡，可將多個影像和文字混合在一起。

Now click on the Advanced box below the image to fine tune the way the image is blended.

現在點擊影像下方的 "高級 "框，對影像的混合方式進行微調。

In this case we'll go with Face Swap, and then write a text But most importantly, we have pretty good continuity between the faces, and the hands almost look totally realistic.

在這種情況下，我們將使用 "臉部交換"，然後寫一段文字，但最重要的是，我們的臉部之間有很好的連續性，而且手看起來幾乎完全逼真。

But when you have imperfections, you can go back to the image prompt, but this time go to Inpaint or Outpaint, and then paint across the parts that are messed up.

但是，當你有瑕疵時，你可以回到影像提示，但這次要轉到 "塗抹 "或 "塗抹"，然後在有瑕疵的部分塗抹。

Tell it what you want to fix, then it will regenerate those parts of the image.

告訴它你想修復什麼，然後它就會重新生成影像的這些部分。

Congratulations, you just built your own artificial Influencer.

恭喜您，您剛剛建立了自己的人工影響者。

You did it!

你做到了

But photos are only the beginning.

但照片只是開始。

Yesterday PikaLabs released an insane video demoing their text-to-video platform.

昨天，PikaLabs 發佈了一段瘋狂的視頻，演示了他們的文本到視頻平臺。

It's closed source, but luckily Stability AI just introduced stable diffusion video last week, and I can only imagine what kind of depraved content you guys will use that for.

它是閉源的，但幸運的是，Stability AI 上週剛剛推出了穩定擴散視頻，我只能想象你們會用它來製作什麼樣的墮落內容。

This has been The Code Report, thanks for watching, and I will see you in the next one.

密碼報告》到此結束，感謝您的收看，我們下期再見。