Midjourney prompts and art direction: visual briefs, concept reviews, and iteration
Use the Midjourney Image entry to prepare subject, light, materials, and composition, review concepts with a checklist, and compare other image models on matched tasks.
The on-site Midjourney product entry currently displays Midjourney Image and keeps its legacy URL. This does not establish V8.1, native 2K output, or a specific SD/HD workflow. This article covers art direction and concept review; the product page provides the start path.
Prepare the visual brief
Define purpose, subject, setting, light, materials, camera framing, and palette. List details to preserve and exclude separately. For references, check the current mode and image requirements first.
Three steps to prepare and inspect
- Fix one scene and an acceptance checklist; record the selected model, mode, and settings.
- Adapt the examples below. Use generation type, quality, and size choices actually offered by the form rather than assuming external commands or legacy parameters work here.
- Inspect subject, composition, details, and exclusions against the brief. Change one major factor per attempt.
Original prompt examples
These remain writing exercises. Widescreen framing, readable text, motion intent, and style are requests rather than guaranteed outcomes. Describing motion in a still-image prompt does not generate a video.
A seaside lighthouse wrapped in thin fog, dusk side light, cinematic widescreen, realistic but not cold, fine materials, low-saturation warm palette. No text, no watermark, no extra people.
A figure in a trench coat at a rain-soaked neon alley, backlight silhouette, cyberpunk illustration, emphasize fabric folds and ground reflections. Half-body framing. No readable shop signs.
Minimal product still life: a ceramic cup on light linen, soft window light, Nordic interior, shallow depth of field. Clean negative space for a brand KV base. No logo, no text.
Aerial dusk city skyline, golden haze, motion intent of a slow push-in with drifting clouds. Produce a stable first-frame still, subject slightly below center.
Rewrite “a somewhat futuristic office” as a Midjourney prompt: subject, time of day, lens, light, materials, palette, hard constraints. Then give a more restrained commercial version.
Three concept workflows
Recipe A: brand mood board
Keep brand elements fixed and vary one lighting or material factor. Save candidates with their prompts. Select by fit, brand consistency, and detail errors rather than a promised completion time.
Recipe B: covers and key visuals
Inspect composition and empty space in the base image before handling headlines and safe zones separately. Compare GPT Image or Nano Banana with the same copy and reference material instead of assuming better accuracy or speed.
Recipe C: game or film concept review
Try day, night, or rain for the same subject while specifying shapes and framing to retain. Record what to keep, remove, and change next. Check character consistency image by image; the model name cannot guarantee it.
Review and compare other models
| Goal | Inspection |
|---|---|
| Mood and direction | Compare with the brief, palette, and light references |
| Product or character features | Check shape, proportions, marks, and repeated elements |
| Copy and layout | Inspect every character and use separate layout tools if needed |
| Editing control | Verify scope and changes to unrelated areas |
| Cost and efficiency | Record actual cost, waiting time, retries, and manual work |
Choose alternatives by supported inputs and observed results instead of assigning mood to Midjourney, fidelity to GPT, and Chinese layout to Nano Banana by default.
Connecting to video and music
Approve the still before selecting a video mode that accepts the required image input. Current cross-model entries include Kling 3.0 Turbo and Google Veo 3.1. Check Suno audio in the actual edit; an image or text plan is not a completed clip.
