Can a Machine Create AI Art That Moves the Human Soul?

https://www.bestprompt.art/blog-2/

WordPress-ready format. Paste into Posts → Add New → Code Editor. Fill SEO metadata into Yoast or RankMath before publishing.

---

What MidJourney Is — and What It Isn’t

MidJourney is a text-to-image AI platform. You type a description; it generates four image variations in roughly 60 seconds. That’s the whole workflow. It runs through Discord — you join their server, type prompts into a channel with a slash command, and the bot responds with images. There’s no canvas, no layers, no history of what you had for breakfast. Just the prompt and the output.

What it is not: a replacement for skilled visual design. It is a generation tool. It produces a starting point, sometimes a very good one, that still requires human judgment about what to keep, what to iterate, what to take somewhere else. The people getting the most out of it treat it like a fast, strange collaborator — useful when it surprises you, frustrating when it doesn’t understand what you meant. Treating it like a magic box that produces finished work on request leads to generic outputs that look exactly like everyone else’s MidJourney outputs, for reasons we’ll get into.

What It Actually Costs in 2025

MidJourney has no free tier as of 2025 — they ended their free trial in 2023. You have to subscribe to generate anything. Current pricing runs four tiers:

Plan	Monthly Price	Annual Price	Fast GPU Time	Key Feature
Basic	$10/mo	$96/yr (~$8/mo)	~3.3 hrs/mo	Commercial rights; no Relax mode
Standard	$30/mo	$288/yr (~$24/mo)	15 hrs/mo	Unlimited Relax mode — most popular tier
Pro	$60/mo	$576/yr (~$48/mo)	30 hrs/mo	Stealth mode (private generation)
Mega	$120/mo	$1,152/yr (~$96/mo)	60 hrs/mo	High-volume production; max concurrency

The Standard plan is the right choice for most new users — Relax mode means you can generate images without eating into your Fast GPU hours, just with slower queue times. That’s usually fine for exploration and iteration. Basic is genuinely limiting: 3.3 hours goes fast if you’re iterating, and once it’s gone it’s gone until next month.

How the Prompt System Works

MidJourney’s diffusion model takes your text prompt and iteratively transforms noise into a coherent image based on patterns learned from its training data. The specificity of your prompt determines how much creative latitude the model has — and therefore how generic or distinctive your output will be.

Here’s the clearest illustration of what that means in practice:

Vague Prompt	Specific Prompt
“A beautiful landscape”	“A serene lake at golden hour, lone weeping willow on the bank, flock of birds silhouetted against a fiery amber sky, mist on the water, wide-angle, cinematic”
Result: generic scenic image indistinguishable from thousands of others	Result: image with specific mood, composition, and lighting that reflects your intent

The parameters that matter most for controlling output:

--ar (aspect ratio) — controls image dimensions. --ar 16:9 for widescreen, --ar 1:1 for square, --ar 9:16 for portrait. This matters enormously for intended use — a social media post needs different proportions than a desktop wallpaper.

--chaos (0–100) — controls how much variation you get across the four initial outputs. Low chaos (0–20) produces similar, coherent results. High chaos (60+) produces wilder, more experimental variations. Useful when you genuinely don’t know what you want and want the model to surprise you.

--stylize (0–1000, default 100) — how strongly MidJourney applies its aesthetic training. Low values follow your prompt more literally. High values produce more “MidJourney-looking” results — often beautiful, often recognizably AI-generated.

--seed — reuse a seed number from a previous generation to reproduce a similar starting composition. Useful when you want to iterate on a specific result without starting from scratch.

Multi-weighting — you can weight specific concepts using double colons: cyberpunk samurai::2 rainstorm::1 tells the model to weight the samurai element twice as heavily as the rain. Useful for complex scenes where the model keeps emphasizing the wrong element.

The Homogenization Problem — and Why It Matters for Your Work

There’s a real risk in AI image generation that most beginner guides skip past. Wharton researchers studying AI-assisted brainstorming found that when a group uses the same AI tool, outputs converge dramatically — in one experiment, 94% of AI-assisted ideas clustered around the same concept, with nine participants independently naming their toy “Build-a-Breeze Castle.” Human-generated ideas were completely unique by comparison.

The same dynamic applies to visual generation. When thousands of people type similar prompts into MidJourney, the outputs share a recognizable aesthetic — what critics have started calling “the MidJourney look.” High contrast, hyper-detailed, a particular kind of dramatic lighting. Technically impressive; increasingly recognizable as AI-generated. If your goal is work that looks distinctive rather than work that looks polished, that’s a problem the tool creates rather than solves.

The counter-move is deliberate specificity combined with niche reference points. “Cinematic lighting” is a phrase in millions of prompts. “Lighting like a 1970s Italian giallo film” is not. “Surreal landscape” is generic. “Surreal landscape in the style of Giorgio de Chirico’s deserted city squares, late afternoon shadows, architectural precision” pulls the model toward a specific visual territory that fewer people are occupying. The tool’s outputs are shaped by the specificity of what you ask for. That’s the skill gap — not the tool itself, but knowing enough about visual language to ask for something precise.

What You Actually Own: The Copyright Situation

This is the most consequential thing most beginner guides get wrong, and the law has been moving fast. Here’s where it stands as of early 2026.

Pure AI-generated output — an image produced by a prompt with no further human modification — is not copyrightable under U.S. law. The U.S. Copyright Office’s January 2025 report (Part 2: Copyrightability) reaffirmed that copyright requires human authorship, and that purely prompt-driven AI outputs don’t meet that standard. This was confirmed by the D.C. Circuit Court in Thaler v. Perlmutter in March 2025, and the Supreme Court declined to review the case in 2026, leaving that ruling in place.

The nuance: human creative decisions applied to AI outputs can be copyrightable. If you substantially modify an AI-generated image — selection, arrangement, significant editing, creative choices about what to keep and what to discard — those human decisions may create copyrightable expression. The Copyright Office evaluates this case by case. There’s no bright line defining exactly how much human input is enough.

Practically: if you are selling AI-generated images commercially, you are selling work that you likely cannot copyright. Anyone can use your prompt (if they know it) and generate identical or near-identical results. Your commercial protection comes from your reputation, your client relationships, and the human creative decisions you layer on top of the generated output — not from intellectual property law covering the images themselves. That’s a meaningful business consideration that the excitement around AI art consistently underemphasizes.

Purely prompt-driven AI outputs don’t meet the human authorship requirement for copyright protection. Your commercial protection comes from what you do with the output, not the output itself.

U.S. Copyright Office, Part 2: Copyrightability (January 2025); Thaler v. Perlmutter, D.C. Circuit (March 2025)

MidJourney vs. the Alternatives

The competitive field has changed significantly in 2025. MidJourney still leads on aesthetic quality and its specific visual style — the outputs feel considered rather than technically correct. DALL-E 3, integrated into ChatGPT, has closed the gap on photorealism and handles text-in-image more reliably. Adobe Firefly is purpose-built for professional design workflows and generates commercially safe images trained on licensed content, which matters if copyright is a concern. Stable Diffusion XL remains the open-source option — free, highly customizable, but requiring technical setup and local hardware or cloud compute.

The practical answer: if you want striking, stylized images with minimal setup, MidJourney. If you need images inside a professional design workflow or want cleaner commercial licensing provenance, Adobe Firefly. If you want maximum control and don’t mind the technical overhead, Stable Diffusion. These aren’t rankings — they’re different tools for different use cases, and the right one depends on what you’re actually making.

Starting Out: What Actually Matters

The fastest way to improve your MidJourney outputs is to develop a vocabulary for describing visual qualities precisely. Not “beautiful” or “cinematic” — these are too common to be useful differentiators. Instead: lighting conditions (golden hour, overcast, bioluminescent, chiaroscuro), surface textures (weathered, iridescent, hammered metal, matte ceramic), compositional references (Dutch golden age portraiture, Brutalist architecture, ukiyo-e woodblock), and mood descriptors that are specific enough to actually constrain the output (melancholic rather than sad; contemplative rather than peaceful).

The Discord community is genuinely useful for this — the shared channels show what other people are generating and, crucially, what prompts produced those results. The gap between prompts that generate remarkable outputs and prompts that generate generic ones is almost entirely vocabulary and specificity. That’s a learnable skill, and watching what others do accelerates the learning curve considerably.

Start with Standard plan. Spend the first week in Relax mode exploring without worrying about GPU hours. Develop ten to fifteen prompt phrases that reliably produce outputs in the visual territory you care about. Then push into Fast mode when you have a specific project that needs iteration speed. The creative floor is low to enter. The ceiling is constrained only by how specifically you can describe what you want to see — and that’s an entirely human skill the tool can’t supply for you.

---

Sources

• Vendr: MidJourney Pricing Guide 2025/2026
• U.S. Copyright Office: Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025)
• Skadden: “Appellate Court Affirms Human Authorship Requirement” — Thaler v. Perlmutter, D.C. Circuit (March 2025)
• Morgan Lewis: “US Supreme Court Declines to Consider Whether AI Alone Can Create Copyrighted Works” (March 2026)
• Penn Today: “Does AI Limit Creativity?” — Meincke, Nave & Terwiesch, Wharton School (2025)
• Jones Day: “Copyrightability of AI Outputs” — analysis of Copyright Office Part 2 report (February 2025)