Midjourney

Midjourney v7 is the best-looking general-purpose consumer image model available. That's not hype — it's what the blind-preference tests on platforms like Artificial Analysis and Hugging Face's image arenas consistently show. v7 wins on "which looks better" roughly 60% of the time against DALL-E 3, around 55% against Imagen 4, and near 70% against most open-source checkpoints.

What does "looks better" actually mean in practice? A few concrete things. Lighting is more believable, especially for portraits and product photography. Skin texture has fewer of the "plastic doll" artifacts that still show up in DALL-E 3. Composition tends to default to more photographically intelligent framing — rule-of-thirds, foreground interest, depth. And for styled illustration, the model has a wider dynamic range of aesthetic references baked in, so a prompt like "editorial fashion magazine illustration in the style of 1970s Push Pin Studios" actually gets you something close to the reference rather than a generic "illustration."

The gap is narrowing, though. Google's Imagen 4 (inside Gemini Advanced) is the closest competitor on photoreal and occasionally beats Midjourney on tricky lighting. Flux 1.1 Pro Ultra, which you can run through Replicate or Fal, is within striking distance on most prompts and costs a fraction. But if you just want a single subscription that gives you "the good one" without having to pick a model or tweak sampler settings, Midjourney still deserves the top pick.

v7 shipped in early 2025 and cleaned up the two things people complained about most in v6: anatomy and text. Hands are now correct in roughly 85% of prompts involving close-up hands — still not perfect, but far better than the v5-era horror shows. Feet, fingers, and eyes have all had similar upgrades. Text inside images is legible in maybe 70% of attempts for short phrases, which is a huge jump from v6's roughly 30%, though Ideogram and GPT Image-1 still win when text is the whole point.

The other quieter improvement in v7 is "style reference" — the ability to drop in a URL or uploaded image as a style seed using the --sref parameter. v6 introduced this, but v7 is dramatically more accurate at actually honoring the reference's color palette, lighting, and brushwork without overfitting to the reference's subject. This is the feature most Midjourney power users would name first as the reason they haven't left.

Character consistency via --cref (character reference) has also improved, though it's still short of what dedicated tools like HeyGen or Runway's character model can do. For keeping the same person across five or six generations of a storyboard, Midjourney is now usable. For 30-frame character consistency at the level a comic or film would need, it's not there yet.

For years the most frequent complaint about Midjourney was that you had to use Discord. You had to join the Midjourney Discord server, navigate to a newbie channel, and type /imagine commands into a chat scrolling with strangers' genitals-and-dragons prompts. Calling that "consumer software" was always a stretch.

That changed in late 2024 when midjourney.com opened up as a full native web app to all paying subscribers. In 2026, this is the default way to use Midjourney, and it's fine. You get a clean gallery view of everything you've generated, a prompt bar at the top, a grid of image variations, and a side panel for editing, remixing, upscaling, and zoom-out. The new "Draft Mode" lets you iterate quickly on low-fi renders before committing GPU time to the full version. "Edit" mode lets you inpaint or extend an image non-destructively.

But the web experience still has rough edges. Organizing generations into folders requires manual tagging. Collaboration features (sharing a project with another subscriber) are primitive. There's no public API, which means power users building pipelines or batch workflows are still working through Discord slash commands or gray-market wrapper services. And if you want to see other people's work for inspiration — the original appeal of the Discord chaos — the community feed on the web is curated-lite and less interesting than just scrolling the real server.

Here's what you need to know about Midjourney's pricing that the landing page kind of buries. Every plan has a "fast hours" budget — that's the time the GPU is actively rendering your images. A single default v7 generation takes roughly 45–60 seconds of fast GPU. So on the $10 Basic plan's ~3.3 hours/month, you get somewhere between 200 and 260 generations per billing cycle before you run out. For someone who's genuinely using it every day, that's not a lot.

The Basic plan also doesn't include "Stealth Mode" — meaning all your generations are publicly visible on your profile and in the community feed by default. Stealth is a Pro ($60/mo) and Mega ($120/mo) feature. If you're making anything client-confidential, you need Pro minimum.

The Standard plan ($30/mo) is usually the right answer for working creatives. 15 hours of fast GPU is enough for most active users, and if you burn through it you still have unlimited Relaxed Mode — slower generations (a few minutes each, queued behind paid traffic) but no hard cap. Pro and Mega are really only worth it if you need Stealth or you're generating at volume for paid client work.

The commercial-use stipulation is also worth reading carefully. If your company makes more than $1 million a year and you're using Midjourney outputs commercially, you need to be on Pro or Mega. Smaller operations can use Basic or Standard. Nobody is going to audit you, but the TOS is clear.

DALL-E 3 inside ChatGPT is Midjourney's most direct competitor for mainstream users. The trade-offs are consistent: Midjourney wins on raw aesthetic quality; DALL-E 3 wins on prompt adherence, text rendering, and conversational iteration. If you prompt "an orange tabby cat wearing a wizard hat holding a sign that says HELLO," DALL-E 3 will give you that exact image. Midjourney will give you something more beautiful that gets two out of three right. For serious visual work, most people end up using both.

Google's Imagen 4 (inside Gemini Advanced, $20/month) is the current sleeper. It's remarkably close to v7 on photoreal, better than Midjourney on text, and included in a subscription you probably already have. The ceiling is still a notch below Midjourney on stylized illustration, but for social-media-ready photos, the gap is real small.

Ideogram 3 (ideogram.ai) is the specialist choice for anything where text matters — logo comps, ad creative, meme images, book covers with readable titles. It has a useful free tier and punches well above its weight for anything graphic-design-adjacent. For hero art? No. For typography-heavy composition? Better than Midjourney.

Adobe Firefly gets an honorable mention because it's the only major generator trained exclusively on licensed data (Adobe Stock + public domain), which gives big enterprises a compliance story Midjourney can't match. Quality is a half-step behind Midjourney and DALL-E, but it's baked into Photoshop and Illustrator.

If you want no paywall and no content restrictions at all, Stable Diffusion via a local install or Civitai is the honest answer. The trade-off is real setup work, variable quality depending on the checkpoint and sampler you pick, and no safety net if you generate something actually illegal.

Pay for Midjourney if aesthetic quality is the single biggest factor in your work. That means designers, art directors, concept artists, high-end marketers, and anyone who's been burned by "AI slop" look in their previous attempts with cheaper tools. The $30/month Standard plan is the sweet spot — enough fast GPU for daily use, no artificial limits, and the web UI is good enough now that it's not the productivity tax it used to be.

Don't pay for Midjourney if you're a casual user who generates maybe a few images a week — DALL-E 3 inside a ChatGPT Plus subscription you probably already have is easily enough. Don't pay for it if you need reliable text in your images; use Ideogram. Don't pay for it if you need API access or batch automation — use Flux or Imagen via an API provider. And don't pay for it as a beginner who isn't sure whether you'll stick with AI image generation; start with Ideogram's free tier or the free Imagen rotations inside Gemini and work up.

Three years in, Midjourney is still the thing it set out to be: the place to go when you want the best-looking AI image without having to think about how to get it. The web app closed the most legitimate complaint from the pre-2025 era. v7 closed most of the anatomy and text gaps that made rivals look like genuine threats. The remaining weaknesses — pricing friction, no free tier, no API — are deliberate product decisions, not oversights, and they're consistent with a company that has chosen to stay a small, independently operated creative studio rather than race to enterprise scale.

If you have $10–30 a month to spend and you care about how your images look, Midjourney is still the one. If you have $0 a month, use DALL-E 3 through a free ChatGPT account, Imagen via Gemini's free rotations, or Ideogram's free tier — and revisit Midjourney when the work you're making justifies the spend.