Most AI image failures I see start the same way. Someone pastes a long string of style adjectives, artist names, and lighting buzzwords, then wonders why the product keeps disappearing or the composition collapses under a real brief. The style layer is not the problem. The foundation is.
I learned this the hard way on a series of e-commerce jobs in 2023 and 2024. I would spend twenty minutes crafting a prompt that sounded sophisticated, generate a batch that looked impressive in the preview grid, and then watch the set fall apart the moment the client asked for consistent product scale or usable negative space. The fix was not more style. The fix was a repeatable structure that forced the model to lock the product and the frame before I ever mentioned mood or aesthetic.
This is the basic prompt structure I still use on almost every commercial job before I add a single style word. It is deliberately plain. That is the point.
Why Style-First Prompts Break Under Pressure
When you lead with “cinematic soft light, shallow depth of field, Wes Anderson color palette, ultra detailed, 8k,” the model treats those instructions as the priority. Product geometry, relative scale, and empty space become secondary. On a lifestyle product shot that needs to survive a 4:5 crop and a headline overlay, secondary is fatal.
I tested this directly. Same product description, same scene, two prompt orders. Style-first produced more immediately attractive frames and more failures on product proportion and text area. Structure-first produced fewer “wow” frames in the first batch and a higher percentage of images that still worked after I applied the client’s crop and copy requirements.
The structure is not glamorous. It is the part that keeps the image usable when the brief gets specific.
The Four-Part Foundation
I write every commercial prompt in this order:
Product and material
Scene and lighting
Composition and constraints
Exclusions
Only after those four blocks are solid do I add style, mood, or reference language. If the foundation is weak, style just makes the failure look more expensive.
1. Product and Material First
Start with the exact object the client is paying for. Name the form, the material, the finish, and any critical proportions. Be concrete. “Ceramic pour-over kettle” is better than “beautiful coffee maker.” “Matte white glaze with subtle texture, precise spout, matching cup” is better than “premium ceramic.”
I treat this block like a product brief, not a poem. The model needs to know what must not change. If the spout length or the handle angle drifts, the entire set becomes unusable for a carousel. Leading with the product forces the model to allocate attention there before it starts inventing atmosphere.
Example opening I actually use:
“Matte ceramic pour-over kettle with precise spout and matching cup, brushed metal filter, soft white glaze with subtle surface texture”
No adjectives about beauty. No mood. Just the object that has to survive the brief.
2. Scene and Lighting Second
Once the product is locked, describe the environment and the light in practical terms. I avoid cinematic language at this stage. “Sunlit kitchen counter, soft morning window light from the left, gentle falloff” tells the model where the light is coming from and how hard it is. “Cinematic volumetric rays” tells the model to prioritize drama over product readability.
For e-commerce and paid-social work I almost always specify:
Surface (counter, table, fabric, etc.)
Primary light direction and quality
Overall color temperature if it matters to the brand
I keep this block short. The goal is readable light that will still work when the image is cropped or has text placed over it, not a lighting diagram for a short film.
3. Composition and Constraints Third
This is the block that saves the most revision time. State the camera position, the framing, and any hard spatial requirements. Negative space is the classic example. If the creative team needs the right third clear for a headline, say it as a constraint, not a preference.
Language I reach for:
“Slight overhead three-quarter view”
“Product centered in the left two-thirds of the frame”
“Clear empty space occupying the right third, no objects”
“Camera height just above the kettle lid”
These instructions are boring on purpose. They reduce the model’s freedom to “improve” the composition by filling empty areas or tilting the camera for drama. When the brief demands usable space, drama is the enemy.
4. Exclusions Last in the Foundation
I list the things that must not appear. Hands, text, busy backgrounds, extra props, extreme wide angles, heavy vignette. This is not a full negative prompt dump. It is a short, high-priority exclusion list that protects the commercial requirements.
I keep this block tight because long negative lists can start to fight the positive instructions. Name the failures you have already seen on this type of job and stop.
Putting the Foundation Together
A complete foundation prompt for the pour-over set looks like this before any style is added:
“Matte ceramic pour-over kettle with precise spout and matching cup, brushed metal filter, soft white glaze with subtle surface texture, on a clean light wood kitchen counter, soft morning window light from the left with gentle falloff, slight overhead three-quarter view, product in the left two-thirds of the frame, clear empty space occupying the right third with no objects, no hands, no text, no busy background, no extreme angles”
That is the entire first pass. I generate a small batch, score it against the brief, and only then decide whether style language is needed.
When and How I Add Style Words
Style comes last, and only if the foundation is already producing usable geometry and space. I add the minimum required to match the brand moodboard. Often that is a short color and atmosphere note: “calm neutral palette, soft natural materials, quiet morning atmosphere.”
I avoid stacked artist names and heavy aesthetic keywords unless the client specifically asked for a reference look. Those terms are powerful and they tend to override the structural instructions I just worked to lock in. If I need a stronger stylistic push, I add it in a second generation pass after I have confirmed the product and frame are stable.
This two-stage approach costs a few extra minutes and saves hours of revision.
A Real Example From a Recent Job
A small skincare brand needed consistent bottle shots with enough negative space for short claims. Early prompts that led with “luxury minimalist, soft glam lighting, high-end product photography” produced beautiful frames where the bottle proportions shifted and the text area kept getting filled with soft shadows or reflections.
I stripped the prompt back to the four-part foundation. Product geometry and material first. Clean surface and controlled light second. Explicit right-side negative space and camera height third. Short exclusion list fourth. Only after a batch showed stable bottles and open space did I add “soft diffused daylight, clean neutral palette.”
The second approach produced fewer Instagram-worthy outliers and more images that survived the actual delivery requirements. The client approved the set with one minor note instead of three rounds of “the bottle looks different in every frame.”
Common Mistakes I Still Catch Myself Making
Even with a structure I trust, I sometimes drift.
Adding mood language too early because the first batch feels “flat.” Flat is often just accurate. Mood can be layered later.
Using vague product descriptions when the client’s reference photo shows specific proportions. Specificity is cheaper than revision.
Treating negative space as a soft request instead of a hard constraint. Models will fill empty areas if you let them.
Overloading the exclusion list until it starts conflicting with the positive instructions. Keep it short and high-priority.
When a batch fails, I do not immediately reach for more style. I check whether the foundation blocks are still clear and correctly ordered.
How This Structure Fits the Rest of the Workflow
This basic prompt is the starting point for almost every Client Brief Lab and Failure Files post you will see here. Once the foundation is locked, I can test model differences, lighting variations, or controlled character consistency without the results being confounded by prompt chaos.
It also makes debugging faster. If the product geometry drifts, I look at block one. If the text area collapses, I look at block three. If the mood is wrong but the structure is solid, I adjust the style layer only. The separation keeps the diagnosis honest.
The Practical Takeaway
Write the prompt the way you would write a production brief for a photographer. Object and material first. Environment and light second. Framing and hard constraints third. Exclusions fourth. Style last, and only after the first three blocks are producing images that could survive a client crop and a headline overlay.
The structure is deliberately plain because commercial work is not a style competition. It is a consistency and delivery competition. The model that looks most impressive in a single frame is rarely the one that gets you through a ten-image set with the fewest revisions.
I use this foundation on nearly every job now. It is the difference between generating until something lucky appears and generating with a clear definition of what “usable” actually means.
Test the structure before you trust the style. That is the only order that has consistently survived the brief for me.
Brands and cases are illustrative. For AI tool features, rules and availability, refer to their official sites.
Leave your thoughts here, too.