Flux vs Midjourney: An Honest Guide for Content Creators
Episode #3: Master AI Image Generation - Flux vs Midjourney Deep Dive! How I recovered from my addiction to... Midjourney.
Welcome to my new newsletter, Cloud Native Creator!
The main topics of this newsletter are content creation, automation and AI tools from the point of view of a part-time content creator on a budget.
This newsletter will summarise everything I learned while writing and marketing my leading newsletter Cloud Native Engineer while working full-time.
In this article, I will introduce you to Flux, a new text-to-image generative model that can be used as an alternative to Midjourney or DALL-E.
Why would I choose a new model if DALL-E is free with ChatGPT and Midjourney is the best model available?
I have quite a few reasons to make the jump.
In this article, we are going to discuss the following:
My journey into text-to-image models
What are the benefits of using Flux over Midjourney?
API Access
Pay-per-use model
Better Typography handling
Open Source license
Local mode
Advanced tools
Reading material
My journey into text-to-image models
Like many content creators reading this article, my journey started like many with DALL-E.
Like you never forget your first love, I will never forget my first experience with DALL-E. That doesn't mean it was a great experience or I want to return to it. :)
I only used DALL-E because it was a free option with ChatGPT.
I used it for a few months with moderate results until I discovered Midjourney.
It was like discovering that you can produce art comparable to the Gioconda, while until then, you have just been playing with crayons.
I fell in love with Midjourney.
I would stay up all night, literally trying to draw all sorts of crazy ideas.
Like trying to draw my own dreams while I was awake.
After a while, I also started to use it to create covers for my newsletter Cloud Native Engineer, and it became a necessity more than just a hobby.
The quality was exceptional; I could only compare it to having your own illustrator available 24/7.
As a creator, I couldn't look for a better tool, but as a Software Engineer, Midjourney fell short of some nice features that I was so keen to use.
At that time (more than 6 months ago), the only way to interface with Midjourney was via a Discord bot.
They implemented a web interface just before I moved to Flux, but still, no API would allow me to integrate with automation tools like Make.com.
I talked in a previous article Revolutionize Content Creation with Make.com: Say Goodbye to Zapier and Taplio about how, as a Software Engineer and part-time content creator I need to automate vast parts of my work. This was really a deal breaker for me.
From a cost perspective, Midjourney was yet another 20$ / month subscription like many. I swear that 2024 was the year of subscriptions for me. Everything was just a subscription away. Every content creator tool was just 15-20$ / month away.
After countless attempts to get rid of this new addition to Midjourney, I tried hard to reduce the number of subscriptions, and I am happy to say that Midjourney is not on my tools list anymore.
However, automation and cost are not the only reasons for choosing Flux over Midjourney.
P.S. All the images of both my two newsletter are generated with Flux.1 Pro, so I believe there is no need to provide here some sample images.
What are the benefits of using Flux over Midjourney?
When doing research for this article, I asked Perplexity.AI (my new research assistant) for a nice definition of Flux
Flux.1 is a state-of-the-art text-to-image AI model with 12 billion parameters, developed by Black Forest Labs, that generates high-quality images from text descriptions using a combination of transformer and diffusion techniques
If what is written above doesn't make sense. Don't worry!
It's a long way to say that, similarly to other models like Midjourney or DALL-E, this model creates high-quality images from a text description.
In the following subsections, I'll describe a few features that make Flux1 stand out from the competition.
API access
As we briefly anticipated in the previous section, the best feature of Flux is that it can integrate with automation tools like Make.com, Zapier or with your own custom code since it provides an API.
As with any other Rest API, you can write code in any language to integrate with Flux.
Pay-per-use model
Instead of trying to squeeze as many images as possible from my monthly subscription, I can relax now and only pay for what I use (a couple of dollars a month).
Of course, the suggestions here might not apply if you generate hundreds or thousands of images per month. Then, a subscription might be more cost-effective. This is probably not what most of you will use Midjourney for.
Cost is always at the root of my decision to use one tool or another.
If you are a part-time content creator like me, you don't have a team of content creators, writers, illustrators, or a huge budget to try all the tools available on the market.
Better Typography Handling
As with any other text-to-image tool, Flux is not great at handling text, but in my own experience, it is still better than what I could achieve with Midjourney.
Don't get me wrong; they all make many mistakes, but Flux is not too bad.
Open-source license
Flux is an open-source model with a commercial version.
The open-source model is available at black-forest-labs/flux: Official inference repo for FLUX.1 models.
There are currently two different open-source versions of Flux, both available to run for free on Hugging Face:
The commercial version is instead called Flux.1 pro and is available in various places like at black-forest-labs/flux-pro – Run with an API on Replicate.
I usually test my prompts on the free versions and then use Flux1.pro for the final version.
The quality of the pro version is much higher but sometimes I get surprised by what I can do with the free models.
Local mode
If you have a computer with a GPU, you can quickly run the open-source versions of Flux locally.
These are the latest hardware requirements:
GPU with at least 24GB of VRAM
A multi-core processor like Intel Core i7 or i9
At least 24GB RAM
Unfortunately, I don't have a GPU-powered computer; otherwise, I'll spend most of my day using Flux. Maybe that's a blessing in disguise.
Advanced tools
Flux integrates with ComfyUI, a popular open-source text-to-image interface, to build and run complex AI image generation workflows.
Additionally, it supports LoRA (Low-Rank Adaptation), an efficient fine-tuning technique that enables customised image generation without complete model retraining.
Imagine training your AI model to generate pictures of yourself on the Moon, under the sea next to the Titanic or with a penguin in the Arctic sea from the comfort of your couch.
I plan to write an article about Flux LoRAs since I want to train a LoRA with my photos to create a professional picture of myself for my LinkedIn profile.
I know that sounds lame, but since what it takes is 10 pictures and a few bucks. Why not give it a try.
After that, I plan to create a model that produces consistent images for my newsletter with the same theme. It is the same technique but uses a bit more advanced LoRAs.