Midjourney V8.1 Prompting Guide: Cinematic Images and Photorealism

This is a working structure, not a sentence you must fill mechanically. If weather is irrelevant, leave it out. If the product is the focus, describe it before the background.

More words do not automatically mean more control. One strong subject and a convincing lighting idea usually give you a better starting point than thirty competing adjectives. Remove details that do not help you judge the image.

For a series, repeat the subject's defining description consistently. Keep the same words for the same traits rather than inventing a new description for every shot.

2. A practical V8.1 parameter map

Parameters belong at the end of the prompt. Select V8.1 explicitly when you need to reproduce a V8.1 workflow; do not assume the platform's current default is the version used in an older tutorial. Availability in Neylora depends on the controls exposed in the generator.

Core controls

Reference controls tied to other versions

Do not assume that adding an older reference flag turns V8.1 into a native character-reference model. Choose a compatible version deliberately. Newer V8 workflows also have an Edit Model; that is a separate tool, not another name for V7 Omni Reference.

Settings not to carry over blindly

• --q / --quality: not a V8.1 quality control.
• Multi-prompt weights using ::: not supported in V8.1. Rewrite the description instead of importing legacy weighting syntax.
• Draft and Turbo: do not assume that modes from another version are available in V8.1.
• Niji: a separate model family, not an anime switch inside V8.1.

Only include controls that your selected version and interface support. A longer parameter string is not inherently a better one.

3. Image prompts, Style Reference and Omni Reference

The same reference image can answer three very different questions:

• Image prompt: “Use this image as broad visual guidance.”
• Style Reference: “Borrow the visual treatment, not the subject.”
• Omni Reference: “Use this particular person or object in a new scene.”

Image prompt: broad visual influence

An image prompt can affect composition, color, subject shapes and mood together. It is useful when you want a new result to resemble a reference overall, rather than isolating only its style or identity.

In Midjourney's text syntax, image URLs go before the written description:

Replace the bracketed placeholder with a real image URL. In an interface with uploads, attach the image using the appropriate reference control instead.

Higher image weight generally gives the image more influence; lower weight gives the words more room. Treat the starting value as something to test, not a universal recipe. Multiple image prompts blend influences rather than preserving each source independently.

Use it for: broad composition, color and overall visual direction. Do not mistake it for a precise face lock.

Style Reference: transfer the look

Style Reference guides palette, medium, texture, lighting and grading. Its purpose is not to copy the reference's person or layout.

A numeric style code is a convenient way to reuse a visual treatment across different scenes. Keep a small catalog of useful codes, including the model version and style weight used with each one.

Write the text prompt around what the new image should contain. “Detailed portrait of a dog” is clearer than “the look of this picture but make it a dog.” Let the reference control communicate the treatment.

Moodboards and personalization have their own controls. Do not assume that --sw is a universal strength slider for every reference type, or that a code will behave identically across model updates.

Use it for: a series of different scenes that should share a palette, texture or grade.

Omni Reference: guide a specific subject

V7 Omni Reference is intended to bring a particular person, object, vehicle or creature into a new image. It uses --oref and --ow, not --sref and --sw.

Start around the default weight and adjust gradually. Very strong reference influence can compete with the new scene or intended style. A reference improves continuity, but intricate details and exact facial likeness can still drift.

Omni belongs to V7. On newer V8 tools, check the separate Edit Model workflow instead of assuming old Omni syntax will work. These options may not all be exposed in every product interface.

Use it for: a subject-led series where reference likeness matters more than staying on the same generation version.

The differences at a glance

Some compatible workflows combine reference types: an image prompt for composition and a style reference for grading, for example. Keep the roles distinct so you know which input to change when the result misses the brief.

4. Consistency: choose the goal before the tool

A. The same character across different shots

There are several kinds of continuity. Similar casting, a shared aesthetic and an identifiable uploaded face are not interchangeable.

If likeness is essential, test the actual reference workflow before committing to a full series. Generate candidates, compare faces and distinctive details, and select carefully. Do not promise a pixel-perfect or biometric copy.

A seed does not solve identity preservation. It is a comparison tool, not a portable container for someone's face. Similarly, personalization and moodboards guide aesthetics; they are not substitutes for a subject-reference tool.

B. The same style or grade

Keep the same style reference, weight and version across the series. Change the scene description while preserving the treatment. A moodboard can be another aesthetic route where available; personalization supplies an additional taste preference.

Record the complete setup, not just the prompt. A changed model or reference setting can alter the look even when the words remain identical.

C. Repeating a shot or testing a change

Fixing --seed helps compare prompt variations under the same conditions. Change one element at a time: lighting, framing or an exclusion. Keep the version and other settings fixed.

Repeatability is not a guarantee of identical output across future sessions or updates. Save the actual image as well as the prompt, seed and settings.

5. Photorealism and a cinematic look

A useful starting point for a photographic image is Raw + restrained Stylize + a concrete photographic description.

• Start with “a photo of…” or “candid photograph” when you want a photograph rather than a broadly realistic illustration.
• Try --raw and a low --stylize value, then raise stylization only if the image needs a stronger aesthetic treatment.
• If the output looks synthetic, test a short exclusion list such as --no cgi, 3d render, plastic skin, airbrushed.
• Judge the result, not the vocabulary. “Photorealistic” is not forbidden, but it is not a guarantee of realism either.

Useful visual language includes close-up, wide shot, low-angle, shallow depth of field, soft bokeh, motion blur, golden hour, rim light, low-key lighting, anamorphic look and film grain.

Camera and film names can suggest an aesthetic, but ISO, focal length, aperture and shutter speed should not be treated as exact photographic controls. If the effect matters, describe it directly: a softly blurred background, compressed perspective, hard flash or a frozen splash.

You can also communicate cinematic intent through a genre or lighting tradition: film noir, restrained low-key cinematography, a wide frame or a controlled teal-and-orange grade. Avoid stacking incompatible references.

A complete cinematic prompt

The example works as a brief because its pieces agree: subject, action, weather and lighting all support the same tense moment. Treat the settings as a starting point and inspect the output before adding complexity.

6. Myths and habits worth dropping

These are practical distinctions, not a blacklist of individual words. A descriptive phrase may be useful in one prompt and noise in another. The question is whether it communicates a visible requirement or merely promises “quality.”

7. Before you generate: a checklist

• Is the shot and main subject clear near the beginning?
• Does the prompt describe one coherent scene rather than a pile of tags?
• For a photographic result, have you tested Raw, restrained Stylize and a photographic description?
• Have you described the optical or lighting effect you actually want?
• Are exclusions short, specific and supported by the selected interface?
• Have you removed legacy parameters that do not work in your version?
• If a particular character matters, have you chosen a subject-reference workflow rather than relying on a seed?
• If only the look must match, have you fixed the style reference and settings?
• Is each reference assigned the right role?
• Is the aspect ratio right for the final placement?
• Have you recorded the version, prompt, references and settings for comparisons?

8. Workflows for real projects

A. From a brief to one cinematic still

Extract the shot size, viewpoint, subject, action, setting, light and grade. Build a focused prompt, choose the aspect ratio, and start with Raw and low Stylize if realism matters. Generate a small set of candidates and refine the weakest part of the brief, not everything at once.

B. Adapt an old prompt to V8.1

Remove unsupported :: weights and --q. Replace generic quality tags with visible requirements. Turn numeric camera shorthand into a description of depth, blur or lighting. Check every reference flag against the selected version rather than copying the old parameter string intact.

C. Build a character sequence

Decide whether you need a recurring character type or recognizable likeness to a specific reference. Test continuity before producing all the shots. Keep the defining description stable, choose a compatible reference workflow, and tell the client what can drift.

D. Create a series with one visual treatment

Fix the style reference or moodboard, version and relevant strength settings. Vary the content while holding the visual language steady. Review the series together; individual images can look good while failing to match as a set.

E. Produce an advertising key visual

Start with product photograph, a clean composition, a deliberate softbox or rim light, and negative space for a headline. Choose the final placement early; 4:5 is a useful starting format for a vertical campaign asset.

Check product geometry, packaging and every letter. Add critical copy and exact brand assets in a design tool when precision matters. A convincing-looking generated logo is not necessarily the correct logo.

Turn the still into a video starting frame

A strong Midjourney image can establish composition, lighting and grading before an image-to-video step in Seedance or Grok. Use the chosen still as the starting frame, then write a separate motion prompt for the video model. Good still-image composition does not automatically describe good motion.

Try the workflow in Neylora's Midjourney generator. The AI chat can also help turn your brief into a structured prompt; review its parameters against the controls you actually have.

Final takeaway

A better Midjourney prompt is a better visual decision, not a longer incantation. Lead with the shot and subject, make the light intentional, describe visible camera effects, and choose references according to whether you need composition, style or a specific subject.

Keep version compatibility in view. That is what turns a collection of attractive generations into a repeatable production workflow.

Adapted for Neylora Academy from the original V8.1 guide. Practical prompting suggestions are starting points, not output guarantees. For current compatibility, consult Midjourney's Version guide, Raw documentation, Omni Reference documentation and Seed guidance. Interfaces and available features can change.