Midjourney Product Photography Prompt: The Material Study Technique

The phrase “raking light from camera-right” is doing four things at once. It specifies the light source position, the angle — low, grazing — the direction relative to the camera, and the resulting shadow geometry. “Dramatic lighting” does zero of those things.

That vocabulary gap is the difference between a Midjourney product shot that looks like a €5,000 studio commission and one that looks like a generated placeholder. The prompt above isn’t complicated. It’s just precise. Here’s what each clause is actually doing.

The Prompt

Why It Works — Breaking Down Each Element

1. “hand-thrown” — a process word, not a noun

“Ceramic mug” gets you a generic mug. “Hand-thrown ceramic mug” activates a different part of Midjourney’s training data — craft pottery photography, where the imperfect rim, slightly uneven walls, and tactile weight communicate handwork. “Hand-thrown” is a material process indicator. It biases the diffusion away from IKEA and toward a studio in East London that charges €400 for a single piece.

You can apply this logic to any object: “hand-forged” (metalwork), “hand-poured” (candles, resin), “thrown on a wheel” (pottery), “hand-spun” (textiles). Process vocabulary changes the aesthetic class of the output.

2. “Matte ash glaze with fine speckle texture” — surface physics, not color

Surface description is the most underused tool in product prompting. “Matte” tells Midjourney how light behaves on the surface — fully diffuse reflection, no specular highlights, no hotspots. “Ash glaze” activates kiln-fire references in the training data. “Fine speckle texture” forces surface detail to appear at close range. Together, they describe material physics, not just appearance. Describe how light interacts with your object’s surface, not just what color it is.

3. “Fingerprint impressions in the clay visible” — the human trace

Most Midjourney product shots look like 3D renders: technically clean, humanly sterile. This one clause breaks that. “Fingerprint impressions” introduces a human trace that the model must render at a surface level, which forces macro detail and communicates craft authenticity. One clause. Completely changes the character of the shot from commercial to editorial.

4. “Raking light from camera-right casting a long shadow” — technical photography vocabulary

“Raking light” is a real photography term — low-angle, grazing light that skims a surface to reveal texture and depth. It is not the same as “dramatic lighting” (too broad), “hard shadows” (describes the result, not the technique), or “side lighting” (missing the angle information). “Camera-right” gives Midjourney a specific direction. “Casting a long shadow” confirms the low angle. This cluster tells the model: one primary light source, shallow angle, hard edge, geometry-revealing. That combination turns a surface into a study.

5. “Aesop aesthetic reference” — brand as visual shorthand

Brand references are shortcuts to entire visual lexicons. Aesop’s product photography vocabulary includes: warm neutrals, natural materials, tight focus on texture, absence of model, editorial restraint, and a fundamental distrust of visual noise. Dropping the brand name tells Midjourney the genre of production value you want. This works better than describing every element individually because the model has absorbed the entire Aesop aesthetic from thousands of training images.

Variations

Variation 1 — Switch the material (brass)

Swap ceramic → brass, linen → concrete. Same vocabulary structure, different material physics. Warmer metals against cool concrete reads industrial restraint.

Variation 2 — Soft window light (Kinfolk editorial)

Replace raking with "soft diffused window light, overcast sky" for a cooler, more Nordic result. Kinfolk instead of Aesop shifts the brand vocabulary — same editorial restraint, different warmth.

Variation 3 — Overhead flat lay

Switch to overhead by removing the lighting direction and adding context elements. The --ar 1:1 square ratio works better for flat lay compositions.

Variation 4 — DALL-E 3 (vocabulary swap required)

DALL-E 3 doesn't accept Midjourney parameters (--ar, --v, --s). Remove them. Replace "raking light from camera-right" with "dramatic low-angle side lighting from the right, hard shadow geometry" — DALL-E 3 responds better to plain-English descriptions than technical photography terms. Replace "Aesop aesthetic reference" with "editorial restraint, warm neutral tones, craft photography aesthetic." The material description stays the same — it works on both models.

Model Compatibility

Midjourney v6 (tested): Fully responsive to brand name references and --s at 250–350. Below --s 200, surface texture descriptions don't hold — the speckle reads as a blur. Keep stylization at 250 minimum.

DALL-E 3: Works with vocabulary swap above. Brand references sometimes don't translate — describe the aesthetic directly: "editorial restraint, warm neutral color palette, craft studio photography." No parameter suffixes.

Stable Diffusion XL: Most responsive to material prompts of any model. Add a negative prompt: (text, watermark, 3d render, blurry background:1.3). Use CFG 8–10 for precision product shots.

Failure Modes

Too many objects: Add a second product and Midjourney splits attention. Material study works when the subject is singular. Raking light + white background drift: "Warm white linen cloth" + raking light can render grey in some seed iterations. Fix: add "warm cream tone maintained" or switch to "warm sand-coloured linen cloth." Low stylization: Below --s 200, the speckle texture description doesn't stick. Keep at 250 minimum. Seed variance: If four runs produce four completely different glazes, lock a good seed with --seed [number] and run variations from there.

Try It

The material study structure — process word + surface physics + human trace + raking light direction + brand reference — works for any object that rewards close looking. Food, jewellery, ceramics, candles, textiles, hardware. The vocabulary changes; the grammar stays the same. What object are you testing? Drop your run in the comments. If the output misses, tell me what it produced and I'll give you the specific vocabulary swap that closes the gap.