If you've just opened Midjourney (or any AI art tool) for the first time, the blank prompt box is more intimidating than it looks. Type "a dragon" and you'll get something — but it won't look like the polished, cinematic images you've seen other people post. The gap almost never comes down to the tool. It comes down to the prompt. This guide covers Midjourney prompts for beginners: the structure professional AI artists use, the settings worth knowing, and a set of starter prompts you can copy, tweak, and learn from.
Everything here also works, with minor tweaks, in other AI art tools like DALL·E, Stable Diffusion (SDXL), and Gemini's image models — the underlying prompt structure is close to universal.
A short prompt like "a cat in a garden" gives the model almost nothing to work with beyond the subject. It has to invent the lighting, the camera angle, the art style, the mood, and the background — and it usually defaults to the most generic, average version of each one. The images you've seen that look "professional" aren't the result of a better tool; they're the result of a prompt that specified five or six of those decisions instead of zero.
Most strong AI art prompts follow a loose formula, roughly in this order:
[Subject + key details], [setting/environment], [lighting], [camera/lens
or art style], [mood/color palette], [extra details like composition
or texture]
You don't need every piece every time, but the more of them you specify, the less the model has to guess. Here's the same subject with zero structure versus full structure:
a woman in a red dress
A woman in a flowing red silk dress standing on a rain-soaked city
street at night, neon signs reflecting in the puddles, cinematic
lighting with a shallow depth of field, shot on 35mm film, moody and
atmospheric, teal and red color palette
The second version reads more like a film still than a snapshot — not because of a magic word, but because subject, setting, lighting, camera, and mood are all spelled out.
Lead with what the image is actually of, then layer on descriptors. "A weathered lighthouse keeper, deep wrinkles, salt-crusted beard, wearing a yellow raincoat" gives the model a clear anchor before you add anything about setting or style.
Environment does a lot of heavy lifting for mood. Compare "in a forest" to "in a misty pine forest at dawn, shafts of light breaking through the canopy." The second version tells the model what time of day, what weather, and what the light is doing.
Phrases like "shot on 35mm film," "wide-angle lens," "macro photography," or "aerial drone shot" push the model toward photographic realism and a specific framing, instead of a generic illustrated look.
If you want a painting instead of a photo, say so: "oil painting," "watercolor illustration," "3D render," "vintage anime style," "isometric digital art." Naming the medium changes far more about the output than most other single words in a prompt.
A closing phrase like "warm golden-hour tones" or "cold, desaturated, cinematic blues" ties the whole image together and is often what separates a good composition from a striking one.
These are ready to paste into Midjourney, DALL·E, Gemini, or SDXL. Swap the bracketed words for your own subject.
A [subject] in [setting], golden-hour lighting, shot on 85mm lens with
shallow depth of field, cinematic composition, warm color palette,
highly detailed
A whimsical [subject] illustrated in a watercolor storybook style, soft
pastel colors, gentle linework, textured paper background
A hyper-detailed macro photograph of [subject], dramatic side lighting,
shallow depth of field, studio background, ultra-realistic textures
An isometric 3D render of [subject/scene], soft ambient occlusion,
pastel color palette, clean minimal background, product-render style
A moody cinematic portrait of [subject], teal and orange color grading,
volumetric lighting, shot on 35mm film, shallow depth of field
If you'd rather browse fully written examples than build every prompt from scratch, PromptiFex keeps a free, fill-in-the-blank library that includes a dedicated Midjourney & AI Art category. Prompts like the Cinematic AI Image Prompt Generator, the Midjourney Wow-Factor Algorithm for SDXL, and Midjourney Movie Magic turn this whole formula into a single question you answer, instead of a prompt you write from a blank page.
Once you get an image close to what you want, don't throw out the prompt — adjust one piece at a time:
This gets you to a good result faster than starting over with an entirely new prompt each time, and it teaches you which specific phrase caused which specific change.
Do I need Midjourney specifically, or does this work anywhere? The prompt structure in this guide — subject, setting, lighting, camera/style, mood — transfers to essentially every AI image tool, including DALL·E, Gemini's image models, and Stable Diffusion (SDXL). Midjourney tends to reward more descriptive, atmospheric language, while some other tools do slightly better with shorter, more literal prompts, but the core formula holds across all of them.
How long should a Midjourney prompt be? Long enough to cover subject, setting, lighting, and style — often one or two sentences. Going much longer than that usually adds redundant adjectives rather than new information, and can dilute the parts of the prompt that actually matter.
What's the fastest way to get better at this? Keep the prompts that produced images you liked, and treat the ones that didn't as a quick diagnosis: was the subject vague, the lighting unspecified, or the style ambiguous? If you'd rather start from prompts other people have already refined, PromptiFex is free to start and lets you customize ready-made, fill-in-the-blank prompts on iOS, Android, and web instead of building every one from a blank page.
Ready to try these prompts yourself?
Browse 468+ ready-made prompts you can customize and use with ChatGPT, Claude, and Gemini, free to start.