ComfyUI vs. Midjourney: Key Pros & Cons of Popular AI Art Tools
How My AIGC Journey Began
My adventure with AI-generated art started Spring 2023 when I first tried Midjourney. Back then, GenAI was still something new and exciting. Being a curious person, I subscribed to Midjourney for a month just to see what all the hype was about.
That was when I created my first AI-generated image. Four images were generated at once, completely random, purely based on my prompt. Midjourney did its thing, and I just let it go with the flow.
But my ideas weren’t just for fun. At the time, I was working at Allstate Protection Plans, and our team needed images for marketing that wouldn’t have copyright issues. We wanted realistic images of accidental appliance damage, which was tough to find in stock photos. So, I thought, why not generate them myself?
My first AI-generated image was inspired by a stock photo Allstate had purchased (image below).
With Midjourney, I generated a new version (image below), selecting the top-left one for further prompt refinement and editing.
Here are the two realistic-style images I created.
Where Midjourney Fell Short
As I got deeper into AI-generated art, I realized Midjourney wasn’t great for fine-tuning small details. Sometimes, I needed images of very specific, rare scenarios, like a broken office chair arm or a damaged game controller. Midjourney, due to its training limitations, struggled to generate these unusual images.
Using Midjourney felt like gambling. You roll the dice with every prompt. Sometimes you get what you want, but often, the results just don’t match up. If you want to tweak small details, you have to start over, and the new images might not even be as good as the last batch. This lack of control made it hard to create consistent, usable images for marketing materials.
That’s when I discovered ComfyUI earlier this year. Unlike Midjourney, ComfyUI is a node-based AI image generation tool, giving designers much more control and flexibility. It works like Lego, where you connect different nodes to create a custom image generation process. In this article, I’ll share my thoughts on ComfyUI from a designer’s perspective, comparing its strengths and weaknesses to Midjourney.
(For more fun ways to use ComfyUI, check out my other article: "Link." If you’re curious about how AI image generation differs on PC and Mac, see "Link.")
Pros of ComfyUI
Ultimate Control
ComfyUI’s node-based system breaks down image generation into clear, adjustable steps. This makes it easy to tweak specific parts of an image, whether it’s the prompt, sampler, ControlNet, or LoRA models. There’s a learning curve, but once you get the hang of it, the flexibility is incredible.
For designers, having this level of control is a game-changer. You can debug issues easily, refine styles, and create exactly what you need.
High Efficiency
ComfyUI allows you to set up workflows that can be reused over and over. Once your process is in place, generating new images becomes much faster.
Powerful Customization
Unlike traditional WebUI tools that are more limited, ComfyUI is super customizable. You can modify and integrate it with other tools like Photoshop, creating a workflow that fits your needs.
The active ComfyUI community has tons of third-party extensions, allowing users to add extra features and models. You can even develop your own nodes to tailor the system exactly how you want.
Perfect for Complex Tasks
If you need to handle large-scale, long-term design projects, ComfyUI is a better choice than Midjourney. It has a higher creative ceiling and more advanced features.
Visual Workflow
Since everything is visually laid out, ComfyUI helps designers, developers, and project managers communicate better. This makes it easier to plan and test new ideas.
Great for Large Images
ComfyUI can generate ultra-wide or high-resolution images without requiring massive amounts of VRAM.
Cost-Effective
ComfyUI is open-source and runs locally, which means no subscription fees. That's a huge plus compared to Midjourney’s paid model.
Cons of ComfyUI
Steep Learning Curve
For beginners, ComfyUI can be overwhelming. You need to understand node-based workflows and have some basic knowledge of AI image generation.
Complex Interface
ComfyUI’s UI isn’t exactly beginner-friendly. It takes time to get comfortable with it, and adding too many nodes can make your workflow messy. Sometimes, I find myself guessing which node to connect next.
Installation Can Be Tricky
Setting up ComfyUI isn’t as simple as clicking “install.” While one-click installation packages exist, you still need some technical knowledge, especially for Mac users.
For example, I spent one hour installing it the first time. Then I realized that running it on Mac requires PyTorch, which failed 2 out of 3 times. I ended up spending six more hours troubleshooting before switching to Stability Matrix’s desktop version. And let’s not even talk about the struggle of installing new models and extensions…
Honestly, as a designer without a coding background, this felt like ComfyUI was forcing me to learn coding. 😅
High Hardware Requirements
ComfyUI runs best with a high-performance GPU and enough system memory. Here’s what you’ll need for smooth performance:
GPU: At least 4GB VRAM (8GB+ recommended). RTX 3060 or better is ideal.
RAM: 16GB minimum, 32GB+ recommended.
Storage: At least 200GB free space (SSD preferred for faster model loading).
CPU: Not super important, but at least Intel 10th-gen i3 (or better).
If you’re working with large models like SDXL, you’ll need a high-end GPU like an RTX 4090.
Example images generated by ComfyUI using the Flux model
Example images generated by ComfyUI using the SDXL model
ComfyUI vs. Midjourney
Swipe Right for Full Chart 👇
Comparison | ComfyUI | Midjourney |
|---|---|---|
Control | ✅ Precise control, fully adjustable | ❌ Randomized, "gacha-style" results |
Ease of Use | ❌ Requires a learning curve | ✅ Easy to use, quick to start |
Style | ✅ Adjustable based on model & LoRA | ✅ Strong artistic style, auto-generated |
Output Size | ✅ Can generate ultra-high-resolution images | ❌ Limited, requires stitching |
Cost | ✅ Free, runs locally | ❌ Subscription required |
Future Wishlist: Making ComfyUI Even Better
✨ UI Improvements for Beginners
Right now, ComfyUI is powerful but not very beginner-friendly.
Could we get toolbars for key functions—like Photoshop has?
A more structured node layout would help new users connect components more easily.
🖥️ Cross-Platform Optimization
ComfyUI should run smoothly on PC, Linux, and Mac, but right now, performance varies.
Cloud-based deployment could be a game-changer. Maybe an official guide on setting it up?
☁️ My Failed Attempt at Running Flux on Google Colab
Flux is a set of high-end AI image models known for amazing quality and flexible styles. Since my Mac isn’t powerful enough, I tried running it on Google Colab.
Paid $10 for Google Colab to get extra cloud computing power.
Spent 10+ hours setting it up.
First few attempts worked fine, then I kept getting errors.
After multiple retries, I gave up and went back to SDXL. 😭
Moral of the story? AI art tools are evolving fast, but sometimes they can be a real pain to set up.
Final Thoughts
ComfyUI is an incredibly powerful AI image generation tool, especially for professional designers who need high levels of customization. The learning curve is steep, but the creative freedom it offers is unmatched.
With AI technology evolving at lightning speed (Flux 2.0, SD 3.5, DeepSeek… what’s next?), I’m excited yet overwhelmed. But as a designer, I’m committed to learning and adapting. Even if it means struggling through complex installs and debugging errors—I’m ready to embrace the AI revolution! 😆
Lastly, a big thanks to my partner Steven Dee for his technical support and different perspectives as a 3D/AI designer. Also, shoutout to my fellow Product Designer Tina Tsung for inspiring this article. You guys rock! ❤️
