'PKM-06, 💎05, 👾03' Obsidian and the torrential flood of information (from AI Art)
Preamble
I realize I talk about without saying much because I think there's simply too much to cover and so I wanted to give a general overview of what I'm doing. I will likely start talking about plugins more in-depth and giving specific examples.
Obsidian Plugins mentioned:
Templater
Dataview
Account Linker
Image Toolkit
Introduction
I've recently felt overwhelmed by the sheer amount of information there is that exists surrounding Stable Diffusion and AI art in part because it's basically trying to understand art history, which is something people spend years dedicating their life to in higher education. Consequently, I have had an immense pressure to optimize and increase my productivity. I was previously spending alot of time doing everything manually and I often had dozens of tabs if not hundreds of tabs that still needed to be processed. There had to be a better way.
AI was the solution or rather more aptly algorithms, automation, and macros. I began to look into various plugins and macros that would increase my workflow. I knew that AutoHotKey is quite powerful and had used it before. However, I often found myself fighting myself trying to use it because I didn't use it often enough to completely memorize all the syntax resulting in my frequently browsing the web for answers. If I stopped using it for a while I would come back and have to learn everything all over again. That's why I decided to use a developed macro program (it's paid and can do everything AutoHotKey can, but is much easier to get going) as well as Obsidian plugins. I'll talk about that app another day, but today I want to give a broad overview of what I'm doing and how automating steps of it will make my data collection much easier.
The state of tools for AICG
Right now, we're in the wild west stage of AI or as a Japanese translation put it, a "Warring States" period between many different developers vying for a piece of the pie. Consequently, there isn't any one application that clearly stands out in terms of managing not only the information surrounding about the AI technology, but also the information that is generated by the AI--Stable Diffusion in this case. There are a number of interesting apps, but because they have a singular focus often with only a few developers if not only one. They are at least now rather limiting. This led me to lean on Obsidian because I've always wanted a holistic solution to information management and to date Obsidian is the closest thing to it.
The tools currently available for managing assets I found are lacking and that's a good thing because it forces me to create a solution instead of simply searching for a solution. What I wanted is a way to keep track what I've generated and in some cases see where those ideas came from. The current iteration of software that exists is either a SaaS or relatively new meaning that the tools available to us are limited. Lexica is arguably one of the more popular services, but they have since stopped logging prompt data and instead are using AI and algorithms (OpenCLIP) to try to reverse-engineer prompts. This results in low resolution data with a lot of fuzziness that can be annoying for people who have a vision, but are unable to describe it to the AI. If you don't believe me, try to use an auto captioning or tagging tool for some of your images for a dataset and see how the output looks. Trash in is trash out is what I've heard and what I've learned when trying to train an embedding. On to Obsidian.
Obsidian is the best solution I’ve found
One challenge with Obsidian is it reminds me of the old days of webdesign before there were frameworks and scripting and everything was mostly done by hand without the ability to create standardized pages that can be filled in by the web engine. The closest thing we have with Obsidian is dataview and templater, but they do have some limitations for all, but the most advanced coders who have a deep understanding of JavaScript. Having said that the notes still must be manually created. Consequently, I resort to macros and python to supplement my note taking and to help me rapidly create new pages and populate them with content in a standardized way. Having the files in plain markdown is such as simple, yet elegant feature of Obsidian as you can interface with your notes directly without the need for a fancy API or duct tape solutions.
Here is an example of some images I took from StylePile to learn more about art history and discover different artists and art styles. Within StylePile that is all you would be able to see--static images. However, with Obsidian I am able to find interesting artists and learn more about them simply by clicking on the image, which will load up their profile note which will provide links to their social media as well as some of their other artworks (This isn't quite working as intended because of some quirks).
This starts to illustrate the real power of Obsidian and how it has the potential to connect everything, both internal and external sources, and can be integrated with your existing knowledge. As a creative, we have to realize that things unrelated today may become related tomorrow and we are in some ways creating a dividend engine for the future by building a vast network of knowledge that is related to us. This isn't just a wiki filled with lifeless information because I am building layers on top of it and instead it's an engine that I can use to leverage my interests. Here is an example: suppose the AI created a very nice looking tree texture that's pretty neat looking. Someone who isn't keeping track of this might simply discard the image and never think about it again. I see the future potential of this tree texture so I make a record of it. If down the road I am trying to make a forested 3d animation and want to add a bit of zest to my trees I can search through my Obsidian library and find that very tree texture improving my animation.
This is all possible without plugins, but plugins like Dataview really do take this to the next level because I can generate a list of all the assets that I've saved and then filter by tags or folders. I don't need to have alot of overhead like Notion users who have to manually curate and ensure everything is properly done or it might break. I just need to create a new page and toss a tag on it and Dataview will do the rest. That means the more content I properly store inside Obsidian, the more powerful it becomes. Historically, I might've had to navigate to 3 or 4 different applications to leverage this same kind of power. I might have a 3d asset manager, an image manager, a document manager and a project manager. Obsidian does all of this--it can now visually prototype ideas with Canvas in the latest release.
Obsidian Canvas
So what exactly can Obsidian do that other services and something like an image gallery can't? For one, with the new Obsidian Canvas feature I can directly build on top of my images and see the direction that I am actually taking.
I can also use it to work on ideas and then put them on a canvas (heh) to generate new ideas from and see new possibilities, combinations and directions that I didn't previously think of
Conclusion
I have a number of other posts coming up that will describe some of my workflow possibly with video. That's what I'm working on at the moment and am quite pleased with just how Obsidian has been improving my workflow. In fact, this blog is written in Obsidian, which allows me to quickly reference some of the pictures shared here.