![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/bb-bh-Stable-Diffusion-e1673308093712.png\")
<\/p>\n<\/p>\n
AI is really having a moment. There\u2019s DALL-E, Lensa, ChatGPT. Your social media feed is probably full of new avatars and AI-generated haiku<\/a>. Naturally, we at Backblaze were intrigued by this brave new world of AI-generated content. The technology has been wildly popular, but is not without controversy, raising questions about intellectual property, copyright law, artist disenfranchisement, possible displacement of jobs, and general fear over the rise of the machines. On the other side of that coin, there\u2019s definitely a place for AI in the future of work and life. So, I wanted to experiment with it.<\/p>\n Let\u2019s start with Stable Diffusion.<\/p>\n Stable Diffusion<\/a> is one of the new text-to-image technologies popping up all over the internet that allows users to input words and phrases and get back amazing pictures created by its deep learning model. What makes Stable Diffusion so interesting is that it has been open sourced to allow anyone to create their own models for text-to-image generation.<\/p>\n Today, I\u2019ll walk through how you can do just that using Backblaze B2 Cloud Storage<\/a>.<\/p>\n After playing with an online instance of Stable Diffusion<\/a>, I sought out content on some more ways to use the AI tool. I found several examples of how to use Stable Diffusion with your own images like this one<\/a> and this one<\/a>. The most common use case for this was taking advantage of AI to create art from a model based on your own face. Sounds cool, right? But what if I also had a bunch more pictures in Backblaze B2 Cloud Storage? Could I do the same thing to create art, graphics, branded images, and more, from my content in the cloud? The answer is a resounding YES.<\/p>\n For me, this was a fun experiment, but we see a number of different ways this set up could be used both individually and for businesses. I started with about 20 images or so as fodder for Stable Diffusion\u2019s algorithm. But, that\u2019s just the beginning.<\/p>\n Let\u2019s say you\u2019re a marketing team at a small company. You could<\/em> use Stable Diffusion\u2019s paid version and get access to hundreds of thousands of random images from Google, but you really only care about analyzing and generating photos that are relevant to your business. So, you run Stable Diffusion in a cloud compute instance and have it analyze a Backblaze B2 Bucket where you store your own library of images, which you\u2019ve probably been collecting for years. Set up that way, you have your own customized AI engine that analyzes and generates only images that are pertinent to your needs, rather than a bunch of images you don\u2019t care about.<\/p>\n In this experiment, I used Google Colab, which worked well for my needs. But for a real implementation, you could use a Backblaze cloud compute partner like Vultr<\/a>. Egress between Backblaze and Vultr is free, so the analysis won\u2019t cost you anything beyond what it costs to use the two services.<\/p>\n This could be hugely useful for marketing teams, but we also see the value for individuals or businesses who want to keep their data private but still take advantage of AI technology. This way, you aren\u2019t serving up images on public sites.<\/p>\n So, how does it all work? Let\u2019s get into it.<\/p>\n What you\u2019ll need:<\/p>\n The first thing you\u2019ll need to do is create an account at Hugging Face<\/a>. Hugging Face is the home of the modern AI community and is where Stable Diffusion lives. In your Hugging Face account, navigate to your Account Settings and go to Access Tokens\u2014we\u2019ll need one of these to allow our environment to use the Stable Diffusion engines.<\/p>\n Now as to the environment, this can be on your own computer, in a virtual machine (VM), really wherever. My favorite (and free) method I found was a Google Colab<\/a> notebook created by GitHub user TheLastBen<\/a> that makes the process so incredibly simple that anyone can do this. The Colab notebook also takes advantage of DreamBooth<\/a>, a Google Research project that provides for incredible detail on the art and images created by a diffusion model. In short, this is the easiest way to get really good looking AI art. You can get started with the Colab notebook here<\/a>.<\/p>\n In the Colab notebook there are a ton of different options and a great step-by-step guide<\/a> that explains them, but I\u2019ll walk you through the basic settings to get going:<\/p>\n Now the pictures: You\u2019ll want at least 20 pictures or so for your AI model to analyze in order to avoid creating a bunch of generic person art or nightmare fuel. So bust out your phone and take some selfies! If you have a friend to throw in two or three full body pictures this will help as well. A few optional tips:<\/p>\n If you\u2019re an iPhone user, you will need to take one extra step here to save your files in JPEG format. You can find a guide for that in this article<\/a>.<\/p>\n As you save your photos, make sure the file names include the name you\u2019re going to use when generating your AI art. For example, my photos were all named troy (1).jpg, troy (2).jpg, troy (3).jpg, etc. This is important so that the AI understands what to call you (or your subject) when generating your images.<\/p>\n Once you have your photos, it\u2019s time to upload them to a Bucket in Backblaze B2 Cloud Storage. You can easily do this in the Backblaze mobile app<\/a> or on the Backblaze website<\/a>.<\/p>\n With your selfies safely in Backblaze B2, make sure you make them accessible on your computer using a tool such as Rclone mount. If you don\u2019t have an account yet, you can check out our guide on how to set up and configure Rclone mount<\/a>.<\/p>\n You might be wondering why you should upload the photos to a Backblaze B2 Bucket and then mount the Bucket so that we can access it locally, rather than just saving the files to a local folder?<\/p>\n The answer is simple. In this example, we\u2019re working with a few images representing a single subject, so you likely won\u2019t have issues working from your local drive. As you further experiment with more subjects and more images of each subject, you\u2019ll likely outgrow your local drive. Backblaze B2 Cloud Storage<\/a> scales infinitely so you won\u2019t have to worry about running out of space.<\/p>\n Now, back to the Colab notebook, hit the play button on Instance Images<\/strong> and click the button that shows up to Choose Files<\/strong>. In the pop up, choose your mounted instance of your B2 Bucket and select the photos.<\/p>\n Once they are uploaded, skip the Concept Images<\/strong> section and click the play button for Training<\/strong>. If you\u2019ve done everything right, you should see some ASCII art like this:<\/p>\n Depending on how many photos you selected, this can take some time. So grab a coffee, go for a walk, listen to a podcast, or perhaps all three.<\/p>\n Once complete, click the Play<\/strong> button under Test the Trained Model<\/strong>. This will launch a temporary instance of Stable Diffusion with your new custom model in Gradio<\/a>, which is an open-source Python library for running machine learning apps. Click the Gradio link generated and we\u2019re ready to start making some AI art.<\/p>\n Again, there are a ton of options and configurations but all you really need to do at this point is enter some text into the Prompt<\/strong> box and click the big Generate<\/strong> button.<\/p>\n Creating prompts for AI art is quickly becoming its own art form. There are tons of resources out there to inspire you, but here are a few prompts I used along with the resulting art.<\/p>\n Pro Tip:<\/strong> You may need to click the Generate<\/strong> button a few times if something looks off. This is totally normal\u2014your new AI friend is learning over time, and it does this by repeating the generation process.<\/p>\n And finally for something fun\u2026<\/p>\n It even has an artist signature! Although I\u2019m not sure who fRny Y is?<\/p>\n So, there you have it. Your very own AI engine, customized to generate versions of your face (or your library of images).<\/p>\n Good luck to all the budding AI artists out there. If you give this a try, we\u2019d love to see your images on social media. You can find us @backblaze on Twitter<\/a>, Facebook<\/a>, and LinkedIn<\/a>. I look forward to seeing what you all create!<\/p>\n","protected":false},"excerpt":{"rendered":" Stable Diffusion: a new tool that introduces a lot of questions about how we work\u2014and a lot of opportunities as well. In this post, we’ll talk about how to generate AI image content from your personal photo archive in Backblaze B2. <\/p>\n","protected":false},"author":147,"featured_media":107750,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"content-type":"","footnotes":""},"categories":[7,434],"tags":[468],"class_list":["post-107742","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cloud-storage","category-featured-1","tag-b2cloud","entry"],"acf":[],"yoast_head":"\nKicking the Stable Diffusion Tires<\/h2>\n
Use Cases for Stable Diffusion<\/h3>\n
Getting Started with Stable Diffusion and Backblaze B2<\/h2>\n
\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Stable-Diffusion-1-e1673309858228.png\")
<\/p>\n\n
Training the Stable Diffusion Model<\/h2>\n
\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Stable-Diffusion-2.png\")
<\/p>\nCreating Your Own AI-Generated Masterpieces<\/h2>\n
Prompt: \u201cPhoto of troy digital painting\u201d<\/h4>\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Troy-1.png\")
<\/p>\nPrompt: \u201cPhoto of troy person digital painting\u201d<\/h4>\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Troy-2.png\")
<\/p>\nPrompt: \u201cPhoto of troy person digital painting asymmetrical headshot smiling\u201d<\/h4>\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Troy-3.png\")
<\/p>\nPrompt: \u201dphoto of troy person hand-drawn cartoon\u201d<\/h4>\n
![\"\"](\"https:\/\/www.backblaze.com\/blog\/wp-content\/uploads\/2023\/01\/Troy-4.png\")
<\/p>\n