Three weeks ago, a client asked me to create concept art for their product launch. Nothing crazy—just 15-20 high-quality images showing their app in various "lifestyle" scenarios.
Simple enough, right?
Except I had a problem: I'd been using Midjourney for months, but lately it was… off. The v6 update made everything look too artistic. Too stylized. My client needed photorealism, not Instagram-filtered fantasy.
So I did what any reasonable designer would do: I went down a rabbit hole.
I spent $200, generated over 500 images, and tested every major AI image generator to find out which one actually delivers on its promises.
The winner? Not what I expected.
The Test: Same Prompt, Three Tools
I started with a control test. One prompt, three generators, zero editing.
The prompt: "A professional woman in her 30s working on a laptop in a modern coffee shop, natural lighting, photorealistic, shallow depth of field"
Midjourney v6 gave me something that looked like a Vogue photoshoot. Gorgeous lighting, perfect composition, but the woman looked like a model, not someone you'd actually see at Starbucks. Her laptop screen was a blurry mess (Midjourney still struggles with screens). Beautiful, but unusable for my client.
Stable Diffusion XL gave me seven fingers. Not kidding. I tried four different checkpoints, tweaked the prompt, adjusted CFG scale—still got anatomical nightmares 60% of the time. When it did work, the image was decent but flat. No depth, no atmosphere.
FLUX Pro nailed it on the second try. Natural lighting. Realistic proportions. The laptop screen actually showed a readable interface. The woman looked like a real person, not a stock photo model. I could've sent it to my client as-is.
That's when I knew I needed to go deeper.
What Nobody Tells You About Midjourney
I've been a Midjourney subscriber since v4. Paid $30/month for over a year. I love the Discord workflow (kind of), I love the community, and I love how pretty everything looks.
But here's the problem: Midjourney has an opinion.
It doesn't just generate what you ask for—it interprets, stylizes, and "improves" your prompt. Sometimes that's great. You get images that exceed your expectations, with perfect color grading and composition you'd never think to specify.
Other times, it's infuriating.
I spent an hour trying to generate a simple product shot: phone on a desk, flat lay, white background. Basic e-commerce stuff.
Midjourney kept adding:
- Moody shadows I didn't ask for
- "Artistic" composition angles
- Background elements despite specifying "white background"
- Color grading that made everything look like a Wes Anderson film
By attempt #23, I got something usable. But I'd burned through 23 fast generations ($6.90 worth of my $30 subscription) for a shot that should've taken one try.
This is Midjourney's blessing and curse: it's too good at making things beautiful, even when you need them simple.
The FLUX Pro Revelation
FLUX Pro launched quietly last year. No hype, no influencer marketing, just... good engineering.
I'd ignored it because it seemed too new. Why risk a client project on an unknown tool when Midjourney works (mostly)?
But after my Midjourney frustrations, I tried FLUX for the same product shot.
First try: perfect.
Flat lay. Clean white background. Product in focus. No artistic liberties. Just exactly what I asked for.
I sat there staring at my screen, slightly annoyed that I'd wasted an hour on Midjourney when FLUX solved it in 8 seconds.
So I stress-tested it. Generated 200 images over three days with increasingly difficult prompts:
- Text-heavy designs (logos, t-shirts, posters)
- Complex scenes with multiple people
- Specific lighting conditions
- Products with reflective surfaces
- Architectural renders
FLUX's success rate? About 85%. First-try usable images.
Midjourney's? Maybe 40%. Often needed multiple attempts and prompt tweaking.
The difference isn't quality—both can produce stunning images. The difference is predictability. FLUX does what you ask. Midjourney does what it thinks looks better.
The Stable Diffusion Reality Check
Let me be blunt: Stable Diffusion XL is incredible if you're technical.
If you know what "CFG scale," "sampling steps," "LoRAs," and "embeddings" mean, SDXL is unbeatable. You have complete control. You can fine-tune models. You can run it locally. You pay nothing except compute costs.
But if you're a designer, marketer, or creator who just wants images without a computer science degree? SDXL is a frustrating maze.
I spent two hours setting up ComfyUI, installing models, and learning the workflow. Then another hour figuring out why my generations kept coming out blurry (wrong VAE). Then another 30 minutes learning that I needed specific "negative prompts" to avoid common artifacts.
By the time I generated my first usable image, I'd invested 3.5 hours and hadn't even touched my actual project.
Compare that to FLUX or Midjourney: create account, type prompt, get image. Done.
SDXL is for hobbyists and engineers who enjoy the tinkering. If you just want results, it's overkill.
The Pricing Trap
Here's where everyone gets it wrong.
Every comparison article lists Midjourney at "$10-60/month" and calls it affordable. But that's a subscription you pay whether you use it or not.
I looked at my Midjourney usage over six months:
- January: 312 images (heavy project month)
- February: 89 images (slow month)
- March: 167 images (moderate)
- April: 34 images (vacation)
- May: 201 images (new client)
- June: 76 images (slow again)
Six months at $30/month = $180 paid.
But here's the thing: at API rates for equivalent quality, I would've spent:
- January: ~$23
- February: ~$6
- March: ~$12
- April: ~$2
- May: ~$15
- June: ~$5
Total: $63.
I paid nearly 3x what I actually used because of subscription economics.
FLUX Pro and SDXL (via APIs) charge per generation. Some months I'd spend more than $30. Most months I'd spend way less. But crucially, I'd only pay for what I used.
When Each Tool Actually Wins
After 500 generations, here's when I'd recommend each:
Use FLUX Pro when:
- You need photorealism
- Text must appear in the image (product labels, posters, t-shirts)
- You want prompt accuracy over artistic interpretation
- You're working on client deliverables with tight specs
- You value time over experimentation
Use Midjourney when:
- You want artistic, stylized images
- Composition and color matter more than literal accuracy
- You're exploring ideas and concepts, not executing specs
- You enjoy the Discord community workflow
- You can afford multiple attempts to get what you need
Use Stable Diffusion XL when:
- You're technical and enjoy customization
- You need specific styles that require model fine-tuning
- You're generating hundreds/thousands of images
- You want local generation for privacy
- You have time to learn the ecosystem
Use Myjourney when:
- You want a simpler, more accessible interface than Midjourney's Discord
- You need a balance between ease of use and quality
- You're looking for a straightforward web-based AI art generator
- You want to avoid Discord workflows entirely
For my client project? I used FLUX for 80% of the images, Midjourney for the hero shots that needed extra artistic flair, and SDXL for some quick concept variations.
The Tool Nobody Mentions
Here's what changed my workflow entirely: I stopped committing to one tool.
Different projects need different generators. Why pay $30-60/month for Midjourney when I only need it occasionally? Why subscribe to anything when my usage varies wildly month to month?
I switched to pay-per-use access for all my AI tools. When I need FLUX, I use FLUX. When Midjourney's aesthetic fits better, I use that. When I need SDXL's speed for iterations, I use that.
And when I want Midjourney-quality images without the Discord friction? I use Myjourney—it's like Midjourney but with an actual web interface. Same quality, none of the command-line nonsense.
No subscriptions. No monthly guilt. Just pay for what I generate.
Services like ARES let you access FLUX, SDXL, Myjourney, and other generators on pure usage-based pricing. Credits never expire. Some months I spend $40, some months $8. It flexes with my actual work.
After my $200 testing spree and six months of subscription waste, this approach makes infinitely more sense.
The Bottom Line
There is no "best" AI image generator. There's only the right tool for the job.
FLUX Pro surprised me by being the most reliable—it consistently delivers what I ask for without artistic liberties. That matters more than I thought.
Midjourney remains stunning for creative work where beauty trumps accuracy.
Stable Diffusion XL is powerful but demands technical investment.
The real insight? Don't marry yourself to one tool. Use what fits the project. And for the love of efficiency, don't pay subscriptions for tools you use inconsistently.
Your wallet will thank you.
Which AI image generator do you use? Have you tried FLUX Pro yet? I'm genuinely curious how others are navigating this chaotic landscape.