Prompt Engineering Guide: Image Formulas and Templates (2026)

Disclosure

This guide contains affiliate links (marked ad) and promotes tools we built (marked (ad, own product)). Kie.ai is a third-party product we promote as an affiliate.

---

This is Module 5 of the AI Income Blueprint. It covers prompt engineering across all four income streams.

---

Why this prompt engineering guide exists

Same tool. Same model. Same price. Two different prompts produce wildly different results. This prompt engineering guide gives you the exact formulas that work for every image type.

The person using “beautiful sunset photo 8K ultra HD” gets generic AI slop. The person using a structured prompt with specific subjects, named colors, directional lighting, and targeted constraints gets sellable work.

Your prompt is the only variable you fully control. Master it and every batch you generate is stronger than the last.

What is prompt engineering for images? Prompt engineering for images is the skill of writing structured text descriptions that produce high-quality output from image generation models. A good prompt specifies subject, composition, lighting, style, and constraints. The difference between a vague prompt and a structured one is the difference between generic output and sellable work.

---

Table of Contents

• The Universal Rules
• Model Selection Guide
• Optimal Prompt Length by Product
• Stock Photo Prompts
• Poster Prompts
• Wallpaper Prompts
• Vector Prompts
• Advanced Techniques
• The NB2 Prompt Generator

---

The universal rules

These are my working rules for every model and product type. They come from my own batches, not from the vendors, and a model’s official guide can differ. Test them on your own subjects before you trust them.

---

Model selection guide

Four models cover everything I make. Each excels at specific output types. Prices are per image and were read on October 7, 2026 from the Kie.ai pricing page and the Recraft API pricing page.

Partner links from this table: Kie.ai (ad). Kie.ai does not list Recraft V4 text-to-image, only two Recraft tools (background removal and upscale), so Recraft V4 runs on Recraft’s own API. Recraft has also released V4.1. Kie.ai now lists Seedream 5 Flash and Seedream 5 Pro as well. The Seedream formulas on this page target Seedream 4.5 and are not re-tested on Seedream 5.

The decision tree

What are you creating?
│
├─ Stock photo?
│  ├─ Has a person → Nano Banana Pro
│  └─ No person → Seedream 4.5
│
├─ Poster?
│  ├─ Colorful/graphic (art deco, pop art, geometric) → Recraft V4
│  └─ Atmospheric/dark (fantasy, cyberpunk, cinematic) → NB2 (illustrative style)
│
├─ Wallpaper?
│  ├─ Flat graphic (ukiyo-e, retro, art deco) → Recraft V4
│  └─ Photorealistic/anime/atmospheric → Seedream 4.5
│
└─ Vector/SVG?
   └─ Always → Recraft V4 (text-to-vector)

---

Optimal Prompt Length by Product

More words is not always better. In my own batches, each model and product type had a sweet spot.

---

Stock photo prompts

Seedream 4.5 Formula (No People)

[Editorial/Professional] photograph of [subject with 5+ concrete nouns].
[Setting with 2-3 surface materials].
[One lighting sentence: source direction + color temperature].
[Dynamic moment: "caught mid-", "steam rising", "just sliced"].
[Camera perspective: bird's-eye, macro, 45-degree overhead, low angle].
No 3D render, no illustration, no CGI, no watermark, no text.

Copy-paste templates:

Food template:

Editorial photograph of [DISH] in [VESSEL] on [SURFACE MATERIAL]. [2-3 GARNISHES/SIDES]. [BACKGROUND ELEMENT]. Single [DIRECTION] [WARM/COOL] directional light casting [SHADOW DESCRIPTION], [DYNAMIC MOMENT]. [ANGLE] perspective. No 3D render, no illustration, no CGI, no watermark, no text.

Workspace template:

Professional photograph of [WORKSPACE TYPE] with [3-4 SPECIFIC OBJECTS ON DESK]. [SURFACE MATERIAL]. [BACKGROUND ELEMENT with depth]. [LIGHTING DIRECTION] from [SOURCE] at [COLOR TEMPERATURE]. [ANGLE] shot, [DEPTH OF FIELD]. No 3D render, no illustration, no watermark, no text.

Nature template:

Editorial photograph of [SPECIFIC SUBJECT] in [HABITAT]. [2-3 ENVIRONMENTAL DETAILS]. [TIME OF DAY] light [DIRECTION] catching [SPECIFIC DETAIL]. [LENS PERSPECTIVE] at [SUBJECT LEVEL]. No CGI, no illustration, no watermark, no text.

Nano Banana Pro (People)

People shots require structured, detailed prompts. The model needs specifics about appearance, clothing, pose, environment, camera setup, and lighting to produce realistic results.

At a high level, production-quality people prompts include:

• Style and mood with a film stock reference
• Subject description with age, build, and natural skin details
• Wardrobe with specific fabric types and fit
• Pose and hands with a behavioral action (not static “smiling”)
• Environment with 4-6 objects and surface materials
• Camera with a real lens and aperture
• Lighting with direction, source, and color temperature
• Constraints split into anatomy, camera, and style blocks

Writing these prompts from scratch takes practice. The structure matters more than any single detail. A well-structured 300-word prompt usually beats a rambling 500-word paragraph.

Simplified people prompt template:

Professional [STYLE] photograph of [AGE/BUILD/ETHNICITY] [GENDER] [ACTION/POSE] in [SETTING]. Wearing [SPECIFIC CLOTHING with fabric detail]. [HAND POSITION with finger count if visible]. [4-5 ENVIRONMENT OBJECTS]. [CAMERA MODEL] with [LENS]mm f/[APERTURE]. [LIGHTING DIRECTION] at [COLOR TEMPERATURE]K. No airbrushing, no plastic skin, no extra fingers, no 3D render, no watermark, no text.

For production-quality people prompts without the learning curve, see The NB2 Prompt Generator section below.

---

Poster Prompts

Recraft V4 Formula (Graphic Posters, My Most-Used Formula)

Large-format [type] wall poster combining [approach A] with [approach B].
[Background with named colors].
[Subject filling 50-80% of canvas, 8+ concrete nouns].
[Element interaction].
[3-5 named colors with coverage percentages].
[Surface treatment].
[Aesthetic thesis: "[Style] meets [Style], [wall-readiness]"]

Key rules:

• Natural flowing prose (no bold headers, no MUST NOT blocks)
• Named colors only (no hex codes)
• “The scene extends beyond all four edges” (full-bleed)
• No camera gear, no lens specs, no film stock

Copy-paste template (Art Deco):

Large-format art deco wall poster combining geometric precision with [MOOD]. [NAMED COLOR] background with [GEOMETRIC PATTERN]. A [SUBJECT] rendered in angular geometric segments fills the lower two-thirds, [DESCRIPTION OF FORM]. Each segment alternates between [COLOR 1], [COLOR 2], and [COLOR 3] with crisp flat fills. The scene extends beyond all four edges. [PRIMARY COLOR] dominates at [X]%, [ACCENT] at [Y]%. Crisp geometric edges, fully flat graphic treatment, no gradients. Art deco [QUALITY] meets contemporary wall art, [WALL-READINESS STATEMENT].

NB2 Formula (Atmospheric Posters)

In my batches, atmospheric posters worked with bold section headers and hex color codes (unlike Recraft).

**STYLE (LEAD):** [Genre + quality description]
**SUBJECT:** [8+ concrete nouns with tactile details]
**COLOR PALETTE:** [3-5 hex codes with metal optimization note]
**COMPOSITION:** [Focal point, full-bleed statement]
**LIGHTING:** [Source, direction, quality, deep shadows]
MUST NOT -- RENDERING: [no DOF blur, no lens flare, no paper texture]
MUST NOT -- STYLE: [no border, no frame, no flat vector, no watermark, no text]

---

Wallpaper Prompts

Recraft Formula (Flat Styles)

Vertical [style] phone wallpaper with [background].
[Subject] rendered in [technique] fills the lower 40-70% of the tall frame.
[Vertical leading lines] exploit the 9:16 format.
[2-3 named colors with coverage].
Top 15% of canvas is [low-complexity zone] for widget compatibility.
No text, no watermark, no border, no flat vector art, no clipart.

Seedream Formula (Atmospheric/Anime)

[Art style description]. [Subject positioned in lower two-thirds of vertical canvas].
[Color and atmosphere details].

[Light source, direction, interaction with scene].

No text, no watermark. [Style constraints].
No flat vector art, no clipart, no icon design.

Critical wallpaper rules (all styles):

• Always 9:16 vertical
• Top 15-40% reserved for widgets (low complexity)
• Subject in the lower half
• Mix 3+ styles per batch

---

Vector Prompts

B&W Icon Formula (6 Elements)

1. "B&W line art icon of [subject + pose/view]"
2. "Simplified to [4-8] geometric shapes: [list each]"
3. "Commercial [clipart/fine/medium] detail level"
4. "Bold recognizable silhouette, legible at 64px"
5. "Black lines on white background, no color fills, no grey tones"
6. "[stroke] uniform strokes, no texture variation, all contours fully closed.
    No gradients, no shadows, no texture, no fills, no photographic elements"

Colored vector formula

Clean flat vector illustration of [subject + view]. Simplified geometric
forms with solid color fills and no gradients. [Size/legibility target].
[Named colors for each element]. Crisp clean edges, professional [icon/design] quality.
Absolutely no gradients, no color transitions, no blending, no shadows,
no texture, no watermark.

I treat the anti-gradient line as mandatory for every colored vector prompt. Without it, gradients creep into a share of the outputs.

---

Advanced Techniques

1. Behavioral Anchors Over Static Verbs

“Caught mid-laugh” tends to give a natural expression. “Smiling” tends to give a mannequin. “Just set down the coffee cup, hand still hovering” beats “sitting at a table.” Temporal actions create life.

2. Concrete Nouns Over Adjectives

“Cast-iron skillet on weathered oak cutting board with linen napkin and crumbled feta” contains five concrete nouns. “Beautiful rustic kitchen scene” contains zero. The nouns carry the quality.

3. Named Colors Over Generic

“Warm amber,” “sage green,” “dusty rose,” “slate charcoal” give the model specific targets. “Nice colors” gives it nothing to work with.

4. Constraints Split by Category

For complex prompts (people, atmospheric posters), split your constraints into groups:

• Anatomy: no extra fingers, no deformed hands
• Camera: no wide-angle distortion, no inconsistent lighting
• Style: no airbrushing, no plastic skin, no 3D render, no watermark

This is more targeted than a single flat list of negatives.

5. Iteration

Your first prompt is never your best. Generate, evaluate, adjust one variable, generate again. In my experience, a few rounds of refinement improve most prompts.

Variables to adjust:

• Swap the camera angle
• Change the lighting direction
• Add or remove environmental objects
• Try a different demographic
• Adjust the dynamic moment

---

The NB2 Prompt Generator

Writing production-quality prompts for people shots is the hardest skill in AI image generation. The structured format, texture language, camera specifications, and lighting descriptions take time to learn.

The NB2 Prompt Generator (ad, own product) handles this complexity for you. Describe what you want in plain language, and it generates a full production-ready Nano Banana Pro prompt with all the structure, technical details, and constraints built in.

Try it free: autokeyworder.com/nb2-prompt-generator (ad, own product). The tool page offers 5 free prompts with no signup.

For Seedream, Recraft, and vector prompts, the templates in this guide and the individual stream modules (Stock Photography, Posters, Wallpapers, Etsy) give you everything you need.

---

Frequently asked questions

How long should an image prompt be? It depends on the product type. In my own batches, stock photos without people worked best at 80-150 words, and people shots needed 250-500 words. Wallpapers did best at 50-80 words, while vector icons needed only 40-80 words.

Do beauty words like “stunning” improve image quality? In my batches they did not. Words like “beautiful,” “stunning,” and “gorgeous” pushed the output toward an over-processed, plastic look. Describe what you want with concrete nouns instead of adjective praise.

What is the anti-text constraint and why is it needed? Adding “no text, no watermark” to every prompt keeps random gibberish text out of your images. Without this constraint, unwanted text artifacts show up in a share of generations.

How much does each image model cost per image? On October 7, 2026: Seedream 4.5 $0.0325, Nano Banana Pro $0.09 (1K/2K) or $0.12 (4K), and Nano Banana 2 $0.06 at 2K, all on Kie.ai. Recraft V4 costs $0.04 per raster image and $0.08 per vector image on Recraft’s own API. Sources: the Kie.ai pricing page and the Recraft API pricing page. Prices change, so check both before you budget a batch.

Should I include camera settings in my prompts? For people shots with Nano Banana Pro, yes. Camera body, lens, and aperture specifications push the model toward photographic rendering. In my tests, camera settings changed little for Seedream objects and Recraft vectors.

---

Quick Reference: All Formulas

← Back to AI Income Blueprint | Next: 30-Day Action Plan →

---

Part of the AI Income Blueprint series. See all modules and the full tools page.