Midjourney Depth Cues: Flat Layers vs Staged Foreground (Prompts)

Eleven and a half inches square, the box for Canopy is built to be judged at 200 pixels wide on a store listing, roughly the size of a postage stamp. The game inside is a forest strategy title for 2 to 4 players with a 90-minute play time.

The first cover Wrenfield Games approved looked fine at full size and like a sticker sheet at thumbnail size. Every tree sat on the same plane, every element was the same value, and nothing overlapped. The art director’s note was one line: “It reads flat because it is flat.”

The Box That Looks Flat at Thumbnail Size

Depth in an illustration is not a mood. It is a set of physical signals: what covers what, how fast things get smaller, how much contrast survives distance, and where the light falls. Remove those signals and the eye has no ordering information, so the image collapses into a pattern.

Board game covers fail this way more often than posters or book covers, because the brief usually arrives as a feature list. Four factions, six terrain types, a fox ranger, a badger engineer, a wooden walkway, a river. A model given that list will place every item where it fits and give each one equal weight, which is precisely how a cover ends up with fourteen objects competing at the same size.

  • At full size: busy but readable, and everyone in the review meeting approves it.
  • At thumbnail size: a green rectangle with noise, no focal point, no silhouette.
  • At shelf distance: indistinguishable from the four boxes beside it.

Wrenfield’s brief also said “cinematic lighting”, which added glow without adding distance. Light alone does not create depth if everything it touches sits on one plane.

Everything on One Plane

Here is the prompt that produced the flat cover, kept as written including the trailing parameters.

board game box art, enchanted forest, strategy game, magical atmosphere,
highly detailed, cinematic lighting, 8k, masterpiece, trending on
artstation, beautiful composition --ar 1:1

What came back: a wide forest scene with roughly twenty identifiable objects spread evenly across the frame. Trees at the left edge were the same height as trees in the centre. A fox stood beside a walkway with no overlap between them. The palette was uniformly saturated, so nothing receded. The top third, where the logo had to sit, was the busiest part of the image.

Every word in that prompt describes quality. None describes space. “Highly detailed” pushes the model toward more objects, and more objects at equal scale is the definition of flat. “Cinematic lighting” produced a warm haze over the whole frame rather than a light source with a direction.

This is the trap with art prompts generally: the vocabulary that sounds professional is a vocabulary of finish, not of structure. If you want a specific arrangement, you have to describe the arrangement.

Three Distances, One Frame

The rewrite describes a camera, then three depth bands with distances and a scale relationship, then what each band does to contrast. It also reserves the logo area as a compositional instruction rather than a post-production problem.

Box illustration for a forest strategy board game, printed at 11.5 inches
square and read at thumbnail size.

Camera: 35 mm lens, height 1.1 m, level with the tabletop, slight
downward tilt.

Foreground, 0.3 to 0.6 m from camera: three carved wooden playing pieces
standing on moss, the nearest one tipped over, tack-sharp, filling the
lower third of the frame.

Midground, 2 to 4 m: a curved timber walkway crossing from lower right to
centre left, four ranger figures mid-stride, two of them partly hidden
behind trunks.

Background, 20 m and beyond: five receding pine ridges, each one about
15 percent lighter and lower in contrast than the ridge in front, mist in
the gaps, the furthest ridge almost flat grey.

Scale contrast: the nearest playing piece is 45 times the height of a
ranger figure.

Light: low sun from camera left at 15 degrees, long shadows crossing the
walkway toward the right.

Composition: the walkway enters at bottom right and leads to a gap in the
treeline in the upper third. Keep the top 20 percent quiet for the logo.

Palette: moss green, birch cream, rust. No blue.

No text, no logo, no border, no frame. --ar 1:1 --style raw

What came back: four variations, all with a legible foreground piece, a walkway that visibly crossed in front of trunks, and a background that faded to near-grey. The thumbnail test passed on three of the four: at 200 pixels the tipped piece still read as a shape and the gap in the treeline still pulled the eye.

The fourth failed for a useful reason. The model placed the walkway behind the foreground pieces instead of in front, which removed the overlap that made the midground legible. Adding one clause, “the walkway passes behind the foreground pieces but in front of the nearest ridge”, fixed it on the next run.

What the Model Reads as Depth

Five cues do most of the work. Each one has a phrase that produces it and a visible failure when it is missing.

CuePhrase that produces itFailure when absent
Overlappasses behind, partly hidden by, in front ofObjects sit side by side and read as cutouts
Scale ratio45 times the height of, one third the size ofEvery element competes at the same weight
Contrast falloff15 percent lighter, lower contrast, almost flat greyBackground pulls forward and flattens the frame
Camera position35 mm lens, height 1.1 m, slight downward tiltPerspective is inconsistent between elements
Shadow directionlow sun from camera left at 15 degreesLight wraps everything evenly, so nothing sits on ground

Scale ratio is the cue most people skip, and the one that fixes the sticker-sheet problem fastest. Saying that the nearest object is 45 times the height of a figure in the midground does more than any amount of “epic scale”, because it gives the model a number to hold while it draws.

Contrast falloff is the second. A background that keeps full saturation will always look like wallpaper behind the subject. Naming a percentage and a direction gives the model a gradient to apply instead of a style to imitate.

Both cues behave differently indoors, where distance is short and light comes from windows rather than a low sun. If your next cover is a room, a corridor, or a cave mouth rather than a treeline, the band structure still applies but the falloff needs smaller steps; our Midjourney scene prompts cover that interior case.

Your Own Depth Recipe

Fill this in for any cover, poster, or key art. The structure travels; only the objects change.

[Format] illustration for [product], printed at [size] and read at
[smallest viewing size].

Camera: [focal length], height [x] m, [tilt].

Foreground, [distance]: [one object], [state], [focus], occupying
[lower third or named area].

Midground, [distance]: [scene element] crossing from [start] to [end],
[number] of [figures], some partly hidden by [object].

Background, [distance]: [number] of [layers], each [percentage] lighter
and lower in contrast than the one in front, [atmosphere] between layers.

Scale contrast: the nearest [object] is [n] times the height of [figure].

Light: [source] from [direction] at [angle], shadows falling [direction].

Composition: [entry point] leads to [focal point] at [position]. Keep
[area] quiet for [logo or type].

Palette: [three colours]. No [excluded colour].

No text, no logo, no border. --ar [ratio] --style raw

Two habits make this reliable. First, describe only the smallest viewing size you care about, because art that survives the thumbnail survives everything larger. Second, regenerate before you reword: Midjourney’s variation is wide enough that the same prompt often solves a placement problem on the second or third run, and rewriting too early hides whether the prompt or the roll was at fault.

When the cover depends on a specific lighting character rather than a specific layout, the vocabulary in our Midjourney lighting guide slots into the light line without touching the rest of the structure.

Wrenfield kept the three-band structure for the expansion box and the player mat, changing only distances and object counts. The art still gets argued about in review, but nobody has said “it reads flat” since the prompt started describing space instead of quality.

Sources & Notes

The three-band structure, the scale-ratio rule, and the contrast-falloff wording are our own framing, reviewed on 2026-10-22. Wrenfield Games, Canopy, the 11.5-inch box, and the 90-minute play time are illustrative examples. Percentage figures for contrast falloff are descriptive prompts, not measurements; Midjourney does not interpret them literally, and results vary by model version. Parameter names and supported values change between versions, so check the current documentation before reusing a parameter string. Lens and distance values are photographic conventions used as composition instructions, not a simulation of real optics.

声明:1.本站所有文章,如无特殊说明或标注,均为本站原创发布。本网站中【文本生成/文生图】使用了DeeSseek人工智能生成内容技术, 算法来源说明:本网站未自行开发、训练或部署深度合成算法,所使用的人工智能算法来自:DeepSeek、Chatgpt、Google Gemini。 2.为了网站加载速度优化,本站预览图片进行压缩处理,所以有些不是太清晰,并不影响AI提示词生图、AI设计等内容。3.任何个人或组织,在未征得本站同意时,禁止复制、盗用、采集、发布本站内容到任何网站、书籍等各类媒体平台。如若本站内容侵犯了原著者的合法权益,可联系我们进行处理。--GUUNN.COM