I’ve used Obsidian for several years for my note-taking. I love the graph view, bidirectional linking, and its plain-text philosophy. For some time, however, something has been nagging me: its primary source of content is typing words, and a lot of them.

Recently, I fixed this by pairing my Obsidian account with a local AI image generator. I’m not talking about dropping in stock photos or copy-pasting content from Midjourney or Leonardo.ai. I mean generating images directly tied to my notes that live locally, privately, and yes, without a subscription.

The result is a PKM (personal knowledge management) workflow that finally feels like it lives in both sides of my brain.

Why visuals matter in your Obsidian note-taking system

The science behind combining images and text in your notes

I am not a purely verbal thinker, despite writing for most of my professional life. When I’m mapping out a complex project, building a character for a story, or trying to internalize an abstract concept, a well-placed image does something that another paragraph or bullet points cannot. This image can anchor the idea and give my brain a handle.

Adding visuals the traditional way, through web searches, stock sites, and screenshots, never felt right to me. It felt less personal than I wanted. I found them to be someone else’s interpretation of the concept, not mine. And managing those image files over time was always a time-waster.

What I needed was a way to generate the image I had in my head, instantly, without leaving my workflow.

How to set up a free local AI image generator on your Mac

Choosing the right Stable Diffusion model for your hardware

Credit: Bryan Wolfe / MakeUseOf

As a primarily Mac user, the tool that made this click was Diffusers, an open-source image-generation app from Hugging Face. It’s free, available for download in the Mac App Store, and incredibly easy to install with no command-line knowledge whatsoever. It runs entirely on my machine and leverages Apple Silicon’s Metal GPU, which means it generates faster than you’d expect from consumer hardware.

When you first open it, you’ll be prompted to download a model, which is the creative engine that powers the image generation. If you’re on an M3 or M4 Mac with 16GB or more of memory, I’d suggest going straight to Stable Diffusion 3 Medium. It produces the best photorealistic results of anything available in the app, and your hardware will handle it comfortably. If you own an older Mac, Stable Diffusion 2.1 is a reliable starting point.

If you’re on Windows, there are several alternatives worth considering. This includes ComfyUI, which offers a node-based visual interface and a one-click installer, and Automatic1111, which offers a more traditional web UI. Both are excellent alternatives, but neither is optimized for Apple Silicon the way Diffusers is. Mac users will get noticeably better performance staying in the Hugging Face ecosystem.

The key requirement for all of these solutions is that they run locally. As such, your notes and ideas remain private. Although you can route them through a cloud service, that defeats the whole purpose of a personal knowledge vault.

How to connect Obsidian and Diffusers: a step-by-step workflow

Using Obsidian’s Templater plugin to build your image prompts

Credit: Bryan M. Wolfe / MakeUseOf

Once Diffusers is running, I needed a way to integrate it smoothly with Obsidian rather than treating them as two separate apps that I went back and forth between. Here’s the workflow I landed on.

First, continue to write your ideas and thoughts on Obsidian as you’ve always done. I always make sure to write at least a paragraph about the concept or node I’m working on. The goal here is for the writing to force me to articulate what the image should capture, which makes the eventual prompt dramatically better. If your prompt is half-baked, the image will be too.

Next, use a template to build your visual prompt. For this, I use Obsidian’s Templater plugin, one of the easiest ones available. If you haven’t installed it, go to Settings -> Community Plugins -> Browse and search for Templater, and install it. It’s free and takes about two minutes to set up.

Once it’s written, create a new folder in your vault called Templates, and inside it, create a new note called Image Prompt Template.

I’m currently writing a murder mystery set on a cruise ship, so I’ve spent a lot of time visualizing key characters. This means documenting the “who” and the “what,” as well as the mood, atmosphere, and style. I also include a ‘negative prompt’ to tell Diffusers what to actively avoid (like blurry limbs or cartoonish styles).

Related

Is Obsidian Really Worth the Learning Curve for Note-Taking?

Is Obsidian the ultimate note-taking tool or just an overcomplicated app?

For my first prompt, I’ve included this language: “Subject: Mood/atmosphere: Colors: Style: hyper-detailed, cinematic. Negative prompt:”

Now, whenever I’m ready to create an image for a note, I can open that note, click inside it, and trigger it with a hotkey. You can set this trigger in Templater’s settings under Template Hotkeys. A small window pops up, and I type “Image” to find my template, then hit Enter. The template adds that structure to my note, and I simply fill in the blanks.

Here’s an example of a prompt that I’ve used recently:

Subject: a retired schoolteacher in her mid-60s, silver hair pulled back loosely, reading glasses pushed up on her forehead, sitting at a kitchen table with a cup of tea

Mood/atmosphere: quietly content, unhurried, the feeling of a Sunday morning with nowhere to be

Colors: warm cream, soft morning light, muted sage green

Style: hyper-detailed, cinematic, photorealistic

For a final step, you want to copy your completed prompt from Obsidian and bring it to Diffusers by pasting it into the Prompt text box at the top of the app. Add your negative prompt in the Negative Prompt box below it. In this example, I used:

Negative prompt: blurry, low quality, cartoon, watermark, text, deformed hands, young, glamorous, dramatic lighting

Once both prompts are filled in, choose Generate. I usually do this two or three times to find the composition I want. Believe me, some of the images that Diffusers generates aren’t sound, so you’ll need to view several variations before you find the one you want. Over time, as you get better at writing prompts, the process goes quicker.

You now need to save your image to an assets folder inside your Obsidian vault. For simplicity, I used “/assets/images/” in my Obsidian vault root. Every time you save a generated image, it goes here. In this example, I named the file “schoolteacher.png.”

The final step is to embed the image at the top of your Obsidian note. In this case, I’d add “![[schoolteacher.png]]”. On first open, it orients me immediately, and now I know what this note feels like before I read a single word.

The creative payoff: what changes when your notes have visuals

The connection I’ve established between Obsidian and Diffusers has yielded several benefits. First, it has changed how deeply I engage with individual notes. There’s a sense of authorship that a blank Markdown file with a bullet list doesn’t carry. It has also changed how I navigate my vault. The Obsidian Kanban and Dataview plugins can surface embedded images in card and table views, which turns a visual scan of my vault into something closer to browsing a magazine than reading a spreadsheet. My graph view shows how ideas connect; my image-forward views show what those ideas are. For creative projects in particular, this can truly be transformative.

I’ll be the first to admit: this setup didn’t necessarily make me more productive in a traditional sense. It didn’t help me write faster or capture more notes. Rather, it made those notes better over the long run. I also wouldn’t consider this a “deep research project,” which Obsidian is known to help many with.

If you’ve ever felt like your notes are too flat, too text-heavy, too disconnected from how you actually think, this is worth the afternoon. Your vault doesn’t have to be a filing cabinet. With a local image generator and a few minutes of prompting, it can feel much more like a mind.