Guide
Imagen 4.0 Prompt Recipes for Developers: Copy-Paste Prompts for Product, Portrait, and Scene Shots
A practical prompt-structure guide with reusable templates and negative-prompt checklists
On this page
Why Prompt Structure Beats Prompt Wording
Most prompt failures are not model failures. They come from a prompt that was never really specified: a vague subject, an unstated camera angle, lighting that the model has to guess. When a prompt is one long sentence of ad-hoc wording, you cannot tell which part caused a bad result — so you cannot fix it.
A structured prompt turns image generation into a debugging problem instead of an intuition problem. Instead of hoping that a nicely phrased paragraph lands, you decide which slots exist and fill them deliberately. If the output is wrong, you can look at which slot was underspecified and change exactly that one.
That structure gives you three practical wins in a development workflow:
- Predictability. A subject described in a fixed slot always appears in the same position of your prompt, so behavior stays consistent as you scale from one test image to a batch.
- Diffable iteration. Prompt versions differ by a line or two, not by a whole rewrite, which makes before/after comparison meaningful.
- Reusable components. A lighting clause that works for product shots can be lifted into portrait or scene prompts without rethinking the whole string.
Ad-hoc wording has a hidden cost: it makes every prompt a one-off. You re-derive the same phrasing each time, and you have no way to know whether a change you made helped or whether the model just got lucky. Structured prompts are assets you can version, review, and hand off.
The rest of this guide uses a fixed skeleton and fills it with copy-paste recipes for product, portrait, and scene shots. Treat the wording as a starting point, not a rule — the structure is what you keep.

The Reusable Prompt Skeleton
The skeleton has six ordered slots. Order matters: put concrete, hard requirements early and stylistic preferences later, so the model anchors on what the image must contain before it decorates it.
[subject] [setting] [lighting] [camera] [style] [constraints]
Slot definitions
- subject — the thing that must be in frame: object, person, or environment. Name count ("a single ceramic mug"), material, and color here.
- setting — where it sits: surface, background, environment, time of day.
- lighting — direction and quality of light: soft window light, hard rim light, overcast daylight, studio softbox.
- camera — how it is framed and shot: lens feel, angle, distance, depth of field.
- style — the rendering intent: editorial photography, documentary realism, clean commercial still life.
- constraints — the things you want excluded or enforced: no text, no logos, no reflections, centered composition.
Filling it in
Start with a one-line brief, then expand it slot by slot:
Brief: a matte black water bottle on a stone counter
subject: a single matte black insulated water bottle, slightly worn finish
setting: on a pale stone kitchen counter, plain neutral wall behind
lighting: soft diffused daylight from the left, gentle shadow falling right
camera: three-quarter angle, 50mm lens feel, shallow depth of field
style: clean commercial product photography, true-to-life color
constraints: no text, no logos, no reflections on the bottle, single bottle only
Join the slots into one string. Keep punctuation simple; commas between slots are enough. If a slot does not matter for a given image, drop it rather than filling it with filler — empty slots are cheaper than contradictory ones.
Treat this skeleton as the contract between your application and the model. Your app builds the string, your tests compare outputs, and your version control tracks changes to individual slots.

Copy-Paste Recipes for Product Shots
Product shots live or die on surface, lighting, and unwanted artifacts. The subject is usually simple; the difficulty is making it look real without adding clutter or text.
Recipe: clean studio product
a single [product] in [color/material], centered on a seamless [surface] background,
soft studio softbox lighting from above-left with a subtle fill from the right,
product photography framing, 85mm lens feel, sharp focus on the product with a gentle background falloff,
clean commercial still life, true-to-life color,
no text, no logos, no watermarks, no reflections of the studio, one product only
Recipe: lifestyle product in context
a single [product] resting on a [surface] in a [room type],
[morning/afternoon] natural light through a nearby window, soft shadows,
eye-level angle, 35mm lens feel, shallow depth of field with the product sharp,
editorial lifestyle photography, muted natural palette,
no text, no logos, no extra objects touching the product, clean background
Recipe: flat-lay product
a single [product] photographed from directly above on a [color] matte surface,
even diffused overhead lighting, no harsh shadows,
flat-lay composition, product centered with generous negative space,
minimal product photography, high clarity,
no text, no logos, no hands, no props beyond the surface
Which slots matter most
- setting and lighting drive realism more than style words. A correct surface and a named light direction do most of the work.
- camera controls whether the product looks like a catalog shot or a lifestyle shot. Change distance and lens feel first when iterating.
- constraints matter more here than in any other category. Text, logos, and stray reflections are the most common product failures, and they are cheapest to prevent in the prompt.
Keep the product description concrete: material, finish, and color. Vague product nouns invite the model to invent detail you did not ask for.
Copy-Paste Recipes for Portraits
Portraits are sensitive to two things: how you describe the face and how you describe the light. Vague faces drift toward generic; unnamed light drifts toward flat.
Recipe: natural-light portrait
a [age range] [gender] person with [hair description] and [expression],
seated near a window in a [room type], soft directional daylight from the side,
chest-up framing, 85mm lens feel, shallow depth of field, eyes in sharp focus,
natural documentary portrait photography, realistic skin texture,
no text, no logos, no sunglasses, one person only
Recipe: studio portrait
a [age range] [gender] person with [hair description], neutral expression, facing camera,
seamless [color] studio background, soft key light from the front-left with a subtle rim light,
head-and-shoulders framing, 100mm lens feel, crisp focus,
clean studio portrait photography, natural skin tones,
no text, no logos, no harsh shadows on the face, one person only
Recipe: environmental portrait
a [age range] [gender] person in [clothing] standing in a [environment],
natural ambient light appropriate to the location, soft shadows,
three-quarter framing at eye level, 50mm lens feel, moderate depth of field,
editorial environmental portrait, realistic color,
no text, no logos, face clearly visible, one person only
Which slots matter most
- subject should describe observable attributes — age range, hair, clothing, expression — rather than identity claims. Descriptions you can verify in the output are easier to iterate on.
- lighting is the single biggest lever on skin realism. Name a direction and a quality (soft, directional, diffused) instead of writing "nice lighting".
- camera sets framing. If faces come out too small or too tight, adjust framing distance before rewriting the subject.
Add "realistic skin texture" when you want photographic results, and keep the negative checklist on hand — hands and background clutter are the usual portrait failure modes.
Copy-Paste Recipes for Scene and Environment Shots
Scene prompts are about composition and depth. The subject is often the space itself, so your job is to tell the model where the camera stands and what is allowed to fall out of focus.
Recipe: wide interior
an empty [room type] with [key furniture] and [material] floors,
daylight from large windows on the [side], soft shadows across the floor,
wide angle from corner height, 24mm lens feel, deep depth of field,
architectural interior photography, neutral color palette,
no text, no logos, no people, no clutter
Recipe: outdoor landscape
a [terrain type] landscape with [foreground element] in the near field,
[time of day] light, low sun casting long shadows,
wide establishing shot, moderate depth of field with the foreground sharp,
realistic landscape photography, natural color grading,
no text, no logos, no people, no buildings
Recipe: lifestyle scene
a [room type] with a person [activity] near [object],
warm ambient light in the late afternoon, soft shadows,
medium shot from eye level, 35mm lens feel, foreground and background separated by depth of field,
lifestyle editorial photography, natural palette,
no text, no logos, face not required, uncluttered composition
Which slots matter most
- camera defines the scene. Distance, height, and lens feel decide whether a room reads as spacious or cramped — change these before changing the description of the space.
- setting should carry the depth cues: foreground element, midground subject, background. Naming all three gives the image layers.
- lighting with a named time of day produces far more believable outdoor scenes than abstract mood words.
Scene prompts also benefit most from negative entries about clutter. If the image feels busy, add exclusions before you rewrite the positive prompt.
Negative Prompt Checklist
Negative prompts are cheap insurance. You do not need a long list for every image — instead, keep a short checklist and copy the entries that match the failure modes you actually see.
Universal entries
no text, no logos, no watermarks, no signatures
no duplicate subjects, no extra limbs, no extra fingers
no blurry faces, no distorted features
no cluttered background, no distracting objects
Product-specific
no reflections of the camera or studio, no harsh hotspots
no props touching the product, no hands
no brand marks, no price tags, no packaging text
Portrait-specific
no sunglasses, no masks, no occluded face
no plastic skin, no over-smoothing
no group of people, one person only
no cropped head, no cut-off shoulders
Scene-specific
no people unless explicitly requested
no floating objects, no impossible geometry
no harsh shadows across key surfaces
no fisheye distortion, no tilted horizon
How to use the checklist
- Start with universal entries. They cover the majority of failures with minimal prompt weight.
- Add category entries only when you see the failure. Piling on every exclusion can fight your positive prompt and flatten results.
- Write exclusions as concrete nouns and defects. "No text" and "no extra fingers" are actionable; "bad quality" is not.
- Mirror your positive prompt. If the positive prompt says "single product", pair it with "no duplicate subjects" rather than assuming one implies the other.
Keep this file next to your prompt templates so every new recipe starts from the same baseline.
Comparing Outputs Before You Ship a Prompt
A prompt that looked good once is not a prompt that is ready for production. Before you wire a template into an application, run a short comparison loop and lock a version.
The loop
- Fix a baseline. Start from one editable template and one representative brief for each image type you ship.
- Change one slot at a time. Adjust lighting, then camera, then constraints — never several at once. If results improve, you know which change did it.
- Generate a small set per variant. A single image can be misleading; a handful per variant shows whether the result is typical or lucky.
- Compare side by side. Judge against your brief, not against each other. Ask: does the subject match, is the lighting right, are the exclusions respected?
- Lock and version. Once a variant wins, freeze it, name it, and store it with the negative checklist it was tested against.
What to record per version
prompt_id: product-studio-v3
category: product
skeleton_version: 1
positive: >
a single matte black insulated water bottle, ...
negative:
- no text
- no logos
- no reflections
notes: softened key light, moved to three-quarter angle
Practical guardrails
- Version the skeleton, not just the string. If you reorder slots, that is a breaking change — record it.
- Keep a small regression set. A handful of briefs per category catches most regressions when you edit a shared slot.
- Re-test after model or slot changes. A prompt tuned for one setup is not automatically tuned for the next.
- Ship the template, not a single lucky prompt. Your application should build prompts from the template, so every request gets the tested version.
When the comparison loop is done, you have something better than a good image: a prompt you can reproduce, review, and improve without starting over. Run your candidates through a single interface, keep the recipes and negative checklists in version control, and promote only the variants that survive the comparison.
Related articles
View all articles →-
Guide
gpt-6.2-sol Getting Started Guide: Capability Boundaries, API Integration, and Token Pricing Essentials
Integration decisions and practical essentials for engineering teams
-
Guide
gpt-6.2-astra Getting Started Guide: Text Model Capability Boundaries, Integration Essentials, and Cost Control
-
Guide
Tripo H3.1 Getting Started Guide: Choosing Among Three Input Modes, Parameters, and Pricing
A practical guide for developers on 3D generation integration decisions and cost control