Midjourney and Nanobanana Prompt Cheat Sheet

Midjourney is opinionated. Give it a short prompt and it will fill in style, lighting, and composition with its own aesthetic judgment. This makes it great for exploration but sometimes frustrating when you want precise control. Understanding its defaults helps you work with them instead of against them.

Midjourney reads prompts front to back and gives more weight to words that appear earlier. Place your most important descriptors (subject, style) at the beginning and less critical details (background, minor elements) at the end. "A knight in silver armor standing in a foggy forest, oil painting style" prioritizes the knight and his armor over the forest details.

The Midjourney documentation explains its parameter system: --ar for aspect ratio, --s for stylization (0-1000, higher = more artistic interpretation), --c for chaos (0-100, higher = more variation between outputs), and --no for negative prompts. These parameters give you mechanical control that complements your text prompt.

A key Midjourney skill: use the /describe command to upload an image and get prompt suggestions that would recreate it. This reverse-engineering tool teaches you which words Midjourney associates with specific visual qualities. It's the fastest way to learn the vocabulary that works best with this model.

Nanobanana takes a different approach. Where Midjourney adds artistic flair by default, Nanobanana aims for faithful interpretation of your description. If you describe a specific scene in detail, Nanobanana tries to render exactly that scene rather than adding its own creative spin.

This makes Nanobanana particularly strong for photorealistic output, product visualization, and any task where accuracy matters more than artistic surprise. When you need the image to match a creative brief precisely, Nanobanana's literal approach is an advantage.

Nanobanana prompts benefit from more detail than Midjourney prompts. Where Midjourney can work well with "a cyberpunk city at night," Nanobanana produces better results with "a cyberpunk city street at night, wet pavement reflecting neon signs, steam rising from manhole covers, a lone figure in a hooded jacket walking toward the camera, 35mm film photography, wide angle lens." The extra specificity gives the model concrete visual anchors to work with.

The Nanobanana platform is gaining popularity for its prompt library and community features. Browsing existing prompts and their outputs is one of the best ways to learn what works. Pay attention to which descriptors produce the biggest visual impact and incorporate them into your own prompts.

Here are proven prompt structures for common use cases, adaptable for both Midjourney and Nanobanana:

Photorealistic portrait: "A 30-year-old woman with freckles and auburn hair, looking directly at camera, soft natural window light from the left, shallow depth of field, shot on Canon 85mm f/1.4, neutral background."

Product shot: "A matte ceramic coffee mug on a walnut wood table, morning sunlight from the right, soft shadows, minimalist composition, white background fading to light gray, commercial photography style."

Fantasy illustration: "A dragon perched on a crumbling stone tower, stormclouds behind, lightning illuminating the scales, oil painting with visible brushstrokes, dark palette with gold and deep blue accents, epic scale."

Architectural visualization: "A modern beach house with floor-to-ceiling glass walls, white concrete and natural wood, golden hour exterior shot, lush tropical landscaping, architectural photography, symmetrical composition."

Vintage photography: "A 1960s diner counter with red vinyl stools and chrome details, Kodachrome color palette, warm tungsten light, slight grain, shallow depth of field on a milkshake in the foreground."

For Midjourney, append parameters: --ar 3:2 --s 250. For Nanobanana, include the technical details directly in the prompt text. Both approaches steer the output toward the same visual goal through different mechanisms.

The Midjourney docs list many parameters, but these are the ones that matter most for daily use:

--ar [width:height] sets aspect ratio. Common values: --ar 1:1 (square, Instagram), --ar 16:9 (landscape, YouTube), --ar 9:16 (vertical, stories/reels), --ar 3:2 (classic photo).

--s [0-1000] controls stylization. Low values (0-100) follow your prompt literally. Default is 100. High values (500-1000) let Midjourney add heavy artistic interpretation. Start at default and adjust based on results.

--c [0-100] controls chaos, meaning variation between the four output images. Default is 0 (all similar). Higher values produce more diverse results. Useful for exploration: set --c 50 to see a wider range of interpretations, then refine your favorite.

--no [item] is the negative prompt. --no text, watermark, blurry removes common unwanted elements. Keep negative prompts short and specific. Long lists of negatives can confuse the model.

--q [.25, .5, 1] sets quality/compute time. Lower values generate faster with less detail. Use --q .5 for rapid iteration and --q 1 for final output.

--v [version] selects the model version. Newer versions generally produce better results, but older versions have different aesthetics. Specify the version when you want consistent style across a project.

Prompt too vague: "A beautiful landscape" gives the model nothing to work with. Fix: add specifics. What kind of landscape? What season? What time of day? What style? "Autumn mountain lake at dawn, golden larches reflected in still water, wide-angle landscape photography, mist hovering over the surface."

Too many competing subjects: "A dragon and a knight and a wizard and a castle and a forest" puts five focal points in one frame. Fix: pick one or two subjects. Put the rest in the background or generate them separately.

Ignoring aspect ratio: Generating a vertical portrait in a square frame wastes half the composition. Fix: always set the aspect ratio to match your intended use before generating.

Over-relying on negative prompts: A negative prompt with 50 items suggests the main prompt isn't clear enough. Fix: improve the main prompt first. Use negative prompts only for specific recurring artifacts.

Not iterating: Your first generation is a rough draft. Vary one element at a time: try different lighting, swap the style reference, adjust the stylization level. The best images usually come from the third or fourth variation, not the first attempt.

Copying prompts without understanding them: A prompt that works for someone else may not work for your use case. Instead of copying blindly, break down why a prompt works (which descriptors drive which visual elements) and apply those principles to your own descriptions.