Midjourney has been the default answer to “which AI image generator should I use?” for three years running. The output is still lovely, especially for portraits, cinematic lighting and anything that needs atmosphere. The friction is everything wrapped around that output.
There’s no free trial. The cheapest plan is $10 a month for about 200 fast generations, and the £30 Standard tier is where most working designers end up. Prompts historically lived in public Discord channels, and while a web editor now exists, the Discord muscle memory stuck. If any of that has you searching for a way out, here’s what’s actually out there.
Be honest about what you’re leaving behind
Midjourney’s real advantage isn’t any single feature. It’s the aesthetic baked into the model. Type “woman in a field” and you get something that looks like a still from a prestige film, without specifying lighting, lens, film stock or colour grade. Most competitors need that scaffolding written into the prompt.
It also has genuinely good creative controls: –sref for style references, –cref for keeping a character consistent across images, and moodboards that let you nudge the model toward a look. Leaving means rebuilding that muscle. Worth knowing before you cancel.
Ideogram is the closest drop-in replacement
If your biggest Midjourney frustration is text that comes out as alien hieroglyphics, Ideogram solves it. Posters, packaging mockups, T-shirt slogans, social graphics with a headline in them: Ideogram gets lettering right far more reliably than anything else at this price point, and it handles short strings of three or four words well enough to use without retouching.
The free tier gives you around ten slow generations a day, which is enough to evaluate properly. Paid plans start at $8 a month. Magic Prompt rewrites your input into something more detailed, which is handy when you’re stuck, and the aspect ratio presets cover the usual suspects without typing flags. Our breakdown of how Ideogram handles typography in AI images covers the limits in more detail.
ChatGPT’s image generator wins on conversation
DALL·E 3 inside ChatGPT is the least intimidating option on this list. You describe a scene in plain English, get four variations, then say “make the jacket green and move the camera lower” and it does. No parameters, no flag memorisation. The selection tool lets you lasso one region and change only that part, which is faster than re-rolling an entire image.
Bundled with ChatGPT Plus at $20 a month, or usable free through Copilot with limits. The trade-off is restraint. Photorealism looks slightly soft compared with Midjourney, there are fewer style knobs to turn, and very specific compositions sometimes get politely ignored. For concepting and quick mockups, though, it’s the least friction you’ll find. There’s a full walkthrough in this guide to generating images inside ChatGPT.
Adobe Firefly is the one your legal team will approve
Firefly is trained on Adobe Stock, licensed content and public domain material, and Adobe offers IP indemnification for enterprise customers. If you’re producing commercial work for a client with a nervous legal department, that matters more than a slightly prettier render.
It’s also already inside the tools you pay for. Generative Fill and Generative Expand in Photoshop are Firefly under the hood, and the same engine appears in Illustrator and Express. Free accounts get 25 generative credits a month; the $9.99 Standard plan adds more and faster processing. The catch is personality. Firefly’s output is competent and a little safe. It rarely surprises you, which is either the point or a dealbreaker.
Stable Diffusion and Flux give you the keys to the building
Open-weight models are a different category entirely. Stable Diffusion XL, SD 3.5 and Black Forest Labs’ Flux run on your own GPU, cost nothing per image beyond electricity, and can be fine-tuned on a LoRA trained from 20 or 30 photos of a product or a person. ControlNet gives you pose, depth and edge control that no subscription tool matches.
The entry price is hardware and patience. SDXL at 1024px runs on 8GB of VRAM; Flux Dev wants 12 to 24GB unless you use a quantised build. ComfyUI is powerful and looks like a wiring diagram. If you’d rather not buy a graphics card, Replicate and Fal.ai host the same models for a few cents per image. Budget a weekend for the learning curve and you’ll never pay per generation again.
The second tier worth trialling
- Leonardo AI hands out 150 free tokens daily, has a solid Phoenix model and a canvas editor aimed at game assets and character sheets.
- Playground is unusually good at poster layouts, stickers and flat graphic design, with a generous free allowance.
- Recraft exports genuine vector SVG files and lets you lock a brand style set, which is rare and genuinely useful for logos and icons.
- Krea runs a realtime canvas where you sketch and watch the render update live, which changes how you ideate.
Several of these sit alongside the bigger names in our roundup of the best AI image generators that actually deliver.
Match the tool to the job
- Text in the image: Ideogram, then Recraft for vector output.
- Commercial client work: Firefly if indemnification matters, Ideogram or Leonardo otherwise.
- Fast iteration on an idea: DALL·E 3 in ChatGPT, because conversation beats parameter tweaking.
- Character consistency across a series: Midjourney’s –cref is still hard to beat; Stable Diffusion with a trained LoRA is the open alternative.
- Volume on a small budget: local Stable Diffusion or Flux, hosted per-image if you don’t own the hardware.
Relearning your prompts is the real cost
Nobody warns you that your Midjourney prompt library becomes half-useless overnight. –ar 16:9 is a dropdown elsewhere. –stylize 250 has no equivalent. The :: weighting syntax doesn’t exist outside Midjourney, and negative prompts move from –no blur into a dedicated field. Style references don’t transfer at all, and seeds are meaningless across models.
The bigger adjustment is tone. Without Midjourney’s built-in polish, prompts need explicit craft words: 35mm, shallow depth of field, soft window light, muted palette. Strip those out and other models default to something flatter and more literal.
A practical test: write six prompts covering your actual workload, run them through each free tier, and compare on the things you care about rather than the demo images on the homepage. That takes an afternoon and saves a year of subscription regret. If you want the surrounding stack sorted too, the AI apps genuinely worth installing covers the upscalers, editors and asset managers that fill the gap between a good render and a finished piece of work.
One last thing worth checking before you commit: read the licence on the tier you’re paying for. Several tools restrict commercial use on free plans, and a few claim rights over generated images in ways Midjourney does not. That clause in the terms is the difference between a cheap alternative and an expensive mistake.