Common mistakes with Midjourney V7 prompt engineering

Parameter errors in Midjourney v7

Midjourney v7 became the default model in June 2025. Many users struggle against the built-in house aesthetic. This aesthetic adds a cinematic flair that ruins documentary realism. You must use the raw parameter to suppress this bias. The model interprets your prompt more literally when you use this specific flag. You should also lower the stylize parameter to the 100 to 150 range. A high stylize value makes the model more opinionated. This causes products to look subtly invented instead of real. If you generate a square image at the wrong aspect ratio and then crop it in post, you lose the intentional composition the model placed in the frame and often decapitate your subject. Use the ar parameter to set the shape before you write the prompt. An aspect ratio of 4:5 works for Instagram, while 16:9 fits desktops. You should use these parameters to fix your renders.

The default stylize value is 100, but the range runs from 0 to 1000. A low value makes the model follow your words closely. A high value makes the model experiment with flair. Use the chaos parameter to explore different takes, but drop the value to zero once you find a direction you like.

Subject and lighting failures

Vague subjects produce bland results because the model averages millions of examples. A prompt like "a woman" creates a statistical blur. You should give the subject four specific descriptors like hair, clothing, expression, and action. This specificity is the cheapest quality upgrade. Lighting provides depth and prevents a flat look. You should specify the source, the direction, and the mood. For instance, use "candlelight from below, low-key, dramatic chiaroscuro" to add dimension. Use "soft diffused light" or "rim light" to define edges. For natural light, try "morning light from the left" or "backlit by setting sun." For color temperature, pick "warm 3200K" or "cool blue twilight." Do not stack seven conflicting modifiers like cinematic, dramatic, atmospheric, and moody. These synonyms add noise instead of signal. Pick two or three modifiers from different categories. One category should be the medium, one the lighting, and one the optical detail. For example, use "35mm film, golden hour, anamorphic lens flare."

Camera settings and model limits

Camera settings change the visual character of your render. Technical terms like 85mm or f/1.8 activate learned professional associations. Avoid contradictory framing instructions like "wide shot close-up." Pick one rung on the shot ladder, such as a medium shot or an extreme close-up. Many users include too many objects or too much text. The model also fails at rendering text longer than three words. For logos or packaging, use a different tool.

The shot ladder moves from extreme wide shot to extreme close-up. An extreme wide shot makes the subject tiny in a vast environment. A full shot shows the whole body from head to toe. A medium shot captures the subject from the waist up. A medium close-up captures the subject from the chest up. A close-up focuses on the head and shoulders. An extreme close-up focuses on eyes or lips. You can also use a top-down shot for cross-sectional views or a low-angle shot to make subjects feel powerful. For wide views, try "bird’s eye view" or "aerial view."

You should use specific lens and aperture data to improve results. A 24mm wide angle with f/8 creates deep focus for architectural shots. An 85mm lens with f/1.4 creates shallow depth of field and soft bokeh for portraits. A 100mm macro lens with f/2.8 reveals textures in close-ups. You can also reference film stocks like Kodak Portra 400. For cinematic looks, mention "anamorphic bokeh" or "T-stop 2.0."

Will the model ever master perfect text placement? Use the seed to keep your results consistent. Write down the seed when you find a result you like. You can make small changes to the prompt while keeping the seed fixed. This makes the comparison fair.