Midjourney

I've been using Midjourney for about eight months now, mostly for client work and personal projects. It's the tool I reach for when I need visuals that look like they were made by a human with actual taste — not just another AI-generated image that looks like everything else out there.\n\nWhat sets Midjourney apart is its eye for composition and mood. When I type "a woman reading in a cafe at golden hour," I don't just get a technically correct image — I get something that feels like a photograph someone actually took. The lighting, the depth of field, the color grading — it all comes together in a way that feels intentional, not random.\n\nThe V6 model was a big jump. Text rendering finally works (I made a poster with actual readable words last month), and photorealism is genuinely convincing. I've shown people images I made and they assumed they were stock photos. The Style Reference feature is a game-changer for brand work — I can upload a client's existing visual identity and get new images that match their aesthetic without hours of prompt engineering.\n\nThe Discord interface is still weird. I know people complain about it, and they're not wrong — it takes a week to feel comfortable typing commands in a chat server to make art. The web app is in alpha and better, but it's still missing some features. You're essentially generating and selecting, not designing — there's no proper canvas or layer system.\n\nAt $10/month for basic access, it's cheap for what you get. The $30/month plan (15 hours of fast generation) is what most working creatives need. The $60/month unlimited tier is overkill unless you're doing high-volume work.\n\nIf you need quick social media graphics, Canva or DALL-E 3 will be faster. If you need brand-safe, legally cleared images, look at Adobe Firefly. But if you care about visual quality and want images that don't look like every other AI output, Midjourney is still the one to beat.

I started using Midjourney in late 2023 when a client needed mood boards for a rebrand project. I'd played with DALL-E before, but the results always felt... corporate. Clean, polished, forgettable. Midjourney was different from the first prompt — the images had texture, atmosphere, something that felt like a human had made aesthetic decisions.\n\nEight months later, it's part of my daily workflow. Here's what I've actually learned.\n\n**Where it genuinely excels:**\n\nConcept art and mood boards are Midjourney's sweet spot. Last month, I needed to show a client three different visual directions for a wellness app. I spent two hours in Midjourney generating maybe 80 images, narrowed it down to 12 strong ones, and presented three distinct aesthetic paths. The client picked one immediately. Without Midjourney, that would have been a week of searching stock sites and mocking up ideas in Photoshop.\n\nThe Style Reference feature changed how I work with brand clients. I upload their existing brand book or a few key visuals, add `--sref` to my prompt, and the output matches their visual language. I made a series of social media images for a coffee brand last week — uploaded three of their product photos as style references, and every image I generated had that same warm, film-grain, slightly desaturated look. No prompt engineering required.\n\nPhotorealism in V6 is scary good. I made a "photo" of a workspace for a blog post — laptop, coffee cup, notebook, plants. Showed it to three people, all assumed it was a real photo. The lighting inconsistencies that used to give away AI images are mostly gone.\n\n**Where it still frustrates me:**\n\nThe Discord interface is a real barrier. I've watched designers try it for the first time and give up after 20 minutes. Typing `/imagine` commands in a chat server to make art feels like using a terminal to write a poem. The web alpha is better, but it launched without upscaling controls and variation options that the Discord bot has.\n\nThere's no proper editing. You generate, you select, you upscale. If you want to change one element — move the coffee cup to the left, change the shirt color — you're regenerating and hoping for the best. Inpainting exists but it's clunky compared to Photoshop's generative fill.\n\nCharacter consistency is still a problem. I tried making a series of images with the same character for a children's book concept. Even with `--cref` (character reference), the face changed between generations. It's better than a year ago, but not reliable enough for narrative work.\n\n**The pricing reality:**\n\n$10/month gets you about 200 images. That's fine for experimentation, not enough for client work. $30/month (15 hours fast time) is what I pay — it's enough for 2-3 client projects per month plus personal work. I've never hit the limit. $60/month is unlimited fast generation, which matters if you're doing high-volume content production.\n\n**Who should actually use this:**\n\nIf you're a designer, art director, or content creator who needs visuals that look hand-crafted, Midjourney is worth the subscription. If you need quick social media graphics with text overlays, you're better off with Canva's AI tools. If you need legally safe, brand-cleared images for a Fortune 500 client, Adobe Firefly is the safer choice (for now).\n\nI've tried Stable Diffusion locally — the control is incredible, but the setup time and hardware requirements make it impractical for client work. DALL-E 3 is easier but the outputs feel generic. Midjourney sits in the sweet spot: high quality, reasonable workflow, outputs that don't look like AI slop.\n\nMy honest take: Midjourney isn't perfect, but it's the only AI image tool I've used that makes me feel like I'm collaborating with something that has taste. That matters more than any feature list.