image-prompt-engineering

name	image-prompt-engineering
description	Prompt engineering for AI images — tool templates, style library, composition, brand palette, artifact workarounds
type	triggered
domain	content
owners	["visual_concept"]
gates	["VISUAL_CONCEPT"]
tech	[]
topic	[]
triggers	["image-prompt-engineering","Промпт-инженерия для изображений","Промпт-инженерия для AI-изображений — шаблоны по инструментам, библиотека стилей, композиция, бренд-палитра, workaround'ы артефактов"]
related	[]
budget_lines	353
schema_version	1

Image Prompt Engineering

When to Use

• When creating AI-generated images for content (Midjourney, DALL-E, Stable Diffusion, Flux).
• When iterating — the result doesn't meet expectations, the prompt needs refinement.
• As part of $visual-brief — the prompt is created based on the visual brief.

Input

Field	Required	Description
Visual brief	✅	From $visual-brief — what should be in the image, mood, style
Brand palette	✅	From $brand-guidelines — colors (HEX), photography/illustration style
Platform	✅	Determines aspect ratio and style
Tool	✅	Midjourney / DALL-E 3 / Stable Diffusion / Flux — different syntax
Mood	⬚	If not specified in the visual brief — determine from the content context
References	⬚	Examples of the desired style — URL or description
Text on image	⬚	If text is needed — specify (or avoid AI text generation)

If the visual brief is not provided — request it or point to $visual-brief. A prompt without a brief = a random result.

Prompt Anatomy

A prompt is built from blocks, from specific to general:

[1. Subject] + [2. Action/Pose] + [3. Environment] + [4. Style/Medium] + 
[5. Lighting] + [6. Color Palette] + [7. Composition/Angle] + 
[8. Mood] + [9. Technical Parameters]

Block	Description	Keyword Examples
1. Subject	Who/what is in the center. As specific as possible.	"young woman in a navy blazer", "abstract geometric shapes", "coffee cup on marble table"
2. Action/Pose	What the subject is doing	"working on laptop", "looking at camera", "floating in mid-air"
3. Environment	Where it is located, background	"modern office", "white studio background", "autumn forest"
4. Style/Medium	How it looks (see library)	"editorial photography", "flat illustration", "3D render"
5. Lighting	Type of light	"soft natural light", "dramatic side lighting", "golden hour"
6. Palette	Colors (tie to brand)	"teal and coral color scheme", "monochrome blue tones"
7. Composition	Angle and framing (see table)	"close-up", "bird's eye view", "rule of thirds"
8. Mood	Emotion / atmosphere	"calm and professional", "energetic and bold", "minimalist and clean"
9. Parameters	Technical (depends on the tool)	--ar 4:5 --v 6.1 --q 2 (MJ), size (DALL-E)

Style Library

Photography

Style	Keywords	When to Use
Editorial	editorial photography, magazine style, high fashion	Image content, brand
Lifestyle	lifestyle photography, candid, authentic, natural	Social media, everyday scenes
Product	product photography, studio lighting, clean background	E-commerce, goods
Documentary	documentary style, photojournalism, raw, unfiltered	Case studies, behind the scenes
Cinematic	cinematic, anamorphic, film grain, shallow depth of field	YouTube thumbnails, hero images
Minimal	minimalist photography, negative space, clean composition	B2B, tech, premium
Aerial	drone shot, aerial view, bird's eye perspective	Landscapes, large-scale scenes

Illustration

Style	Keywords	When to Use
Flat	flat illustration, vector style, clean lines, solid colors	UI, infographics, tech
Isometric	isometric illustration, 3D isometric, technical	Diagrams, processes, tech
Watercolor	watercolor painting, soft edges, flowing colors	Lifestyle, wellness, premium
Line art	line drawing, ink illustration, hand-drawn	Minimalism, education
Retro/Vintage	retro illustration, 70s style, vintage poster	Entertainment, nostalgia
Cartoon	cartoon style, character design, playful	Humor, youth, mascot
Geometric	geometric abstract, shapes, bauhaus style	Tech, modern, branding

3D and Digital

Style	Keywords	When to Use
3D Render	3D render, octane render, cinema 4D, glossy	Tech, product, hero images
Clay/Soft 3D	clay render, soft 3D, pastel, rounded shapes	Friendly tech, SaaS, startup
Glassmorphism	frosted glass, translucent, glassmorphism	UI style, modern tech
Neon/Cyberpunk	neon lights, cyberpunk, futuristic, dark	Gaming, nightlife, tech events
Paper craft	paper cut art, layered paper, shadow depth	Creative, craft, eco

Composition and Angles

Technique	Keywords	Effect
Close-up	close-up shot, macro, detailed	Focus on details, emotion
Medium shot	medium shot, waist-up, half-body	Balance of subject and context
Wide shot	wide shot, full-body, establishing shot	Context, environment
Bird's eye	bird's eye view, top-down, flat lay	Overview, flat lay photos
Low angle	low angle shot, worm's eye view	Power, dominance
Rule of thirds	rule of thirds composition, off-center	Dynamics, professionalism
Symmetry	symmetrical composition, centered	Order, premium
Negative space	lots of negative space, minimalist	Room for text, cleanliness
Dutch angle	dutch angle, tilted, dynamic	Tension, energy
Over the shoulder	over the shoulder shot, POV	Engagement, perspective

Prompt Structure by Tool

Midjourney

[subject], [action], [environment], [style], [lighting], [palette], [composition], [mood] --ar [ratio] --v 6.1 --q 2 --s [stylize 0-1000]

Parameter	Description	Values
--ar	Aspect ratio	1:1, 4:5, 16:9, 9:16
--v	Model version	6.1 (current)
--q	Quality	1 (standard), 2 (high)
--s	Stylization	0 (strictly follows prompt) — 1000 (highly artistic)
--no	Negative prompt	--no text, watermark, blur
--chaos	Variance	0 (predictable) — 100 (chaotic)

DALL-E 3

Descriptive text in natural language. Detailed description of the scene, style, mood.
Size: 1024x1024, 1792x1024, 1024x1792.

Features:

• Accepts natural language (not tags).
• Describe the scene entirely, like a story.
• No negative prompt — instead, explicitly state "without text", "no people".
• Better at handling text in images (but not perfect).

Stable Diffusion / Flux

[subject], [action], [environment], [style], [lighting], [palette], [quality tags]

Negative prompt: [what to exclude]
Steps: 30-50, CFG: 7-12, Sampler: DPM++ 2M Karras

Features:

• Comma-separated tags, quality tags at the end: masterpiece, best quality, 8k.
• Negative prompt — a separate field, critically important.
• More control via ControlNet, LoRA, img2img.

Tying to the Brand Palette

How to embed brand colors from $brand-guidelines:

Method	Example	Reliability
Direct HEX Specification	color palette of #1A73E8 and #FFC107	Medium (AI interprets approximately)
Color Description	deep ocean blue and warm amber tones	High (AI understands descriptions better)
Mood Reference	corporate blue and gold, professional	High
Palette Restriction	monochrome using only shades of teal	High
Negative Control	--no red, green, warm colors (MJ)	High

Best approach: color description + negative exclusion of unwanted colors.

Typical AI Issues and Workarounds

Issue	Symptom	Workaround
Hands / fingers	6 fingers, deformation	Hide hands: "hands behind back", "hands in pockets", "close-up of face only"
Text on image	Garbage text, distorted letters	Add to negative: --no text, letters, words. Add text in an editor. For DALL-E 3: you can try, but verify
Faces	Uncanny valley, asymmetrical	Specify: natural facial features, photorealistic portrait. Or turn away: back view, silhouette
Symmetry	Asymmetrical objects	Add: perfectly symmetrical, mirror symmetry
Too literal	AI interprets a metaphor as an object	Add: artistic interpretation, conceptual
Oversaturation	Colors are too bright, acidic	Add: muted tones, desaturated, soft palette. Negative: --no oversaturated, neon
Too many details	Chaotic composition	Add: minimalist, clean, simple composition, negative space
Wrong style	AI doesn't understand the style	Use reference: in the style of [specific style], add 2-3 style tags
AI artifacts	Blurry areas, compression artifacts	Upscale, inpaint problem areas, or regenerate with --q 2

Protocol

Step 1 — Context Gathering

1. Get the visual brief ($visual-brief) — what to depict, mood, style.
2. Load the brand palette ($brand-guidelines) — colors, style.
3. Determine platform → aspect ratio.
4. Determine the generation tool.

Step 2 — Prompt Assembly

1. Fill all 9 blocks of the prompt anatomy.
2. Choose a style from the library.
3. Choose composition/angle.
4. Embed brand palette (description + negative).
5. Add technical tool parameters.

Step 3 — Negative Prompt

1. Base set: ugly, blurry, low quality, watermark, distorted, deformed.
2. Add contextual exclusions: unwanted colors, objects, styles.
3. Add workarounds (text, hands — if relevant).

Step 4 — Variants

1. Create 3 variants with different focuses:
   • Variant A — primary (exactly to brief).
   • Variant B — different style (alternative medium).
   • Variant C — different composition (different angle).
2. If a version for another tool is needed — adapt the syntax.

Step 5 — Pre-generation Checklist

Before generating, verify:

1. Is the subject specific? (Not "person", but "young woman in navy blazer, short dark hair")
2. Is the style defined? (Not "beautiful", but "editorial photography, soft natural light")
3. Does the palette match the brand?
4. Is the aspect ratio correct for the platform?
5. Is the negative prompt adequate?
6. No request for text (if the tool handles it poorly)?

Validation (Quality Gate)

A prompt is considered ready if:

• All 9 anatomy blocks are filled (or consciously omitted)
• Style is chosen from the library and matches $brand-guidelines
• Palette is tied to the brand (description + negative)
• Aspect ratio matches the platform
• Negative prompt includes the base set + contextual exclusions
• Created ≥ 3 variants with different focuses
• Pre-generation checklist is passed
• Prompt is adapted for a specific tool (syntax, parameters)

Handoff

Consumer	How it's used
$visual-brief	Prompt as part of the visual brief
Visual Concept (agent)	Generates images from prompts
$content-review-checklist	Reviewer checks generation results (item #25 — AI artifacts)
$content-release-gate	AI prompts attached to the publication package

Integration with other skills

Skill	Interaction
$visual-brief	Source: WHAT to depict → prompt describes HOW
$brand-guidelines	Source: palette, photo/illustration style
$platform-visual-specs	Source: sizes and aspect ratios
$moodboard	Source: mood and style references
$content-review-checklist	Consumer: verifies results for AI artifacts

Anti-patterns

Mistake	Why it's bad	How to do it right
"Beautiful image of marketing"	Too abstract, AI doesn't know what to generate	Specific subject: "woman presenting data on screen in modern office"
3-word prompt	Insufficient context	Fill all 9 anatomy blocks
200-word prompt	AI loses focus, erratic result	40-80 words — optimal for MJ/SD
Ignoring negative prompt	Artifacts, text, watermarks	Always add base + contextual negative
Requesting text on image	AI handles text poorly (except DALL-E 3)	Add text in an image editor
Brand colors via HEX	AI interprets approximately	Describe color with words + negative for unwanted
One prompt without variants	No choice, first result might not fit	≥ 3 variants with different focuses
Copying prompts between tools	Different syntax — MJ ≠ DALL-E ≠ SD	Adapt for each tool

Aspect Ratios by Platform

Platform	Format	Ratio	MJ Parameter
Instagram Feed	Square	1:1	--ar 1:1
Instagram Feed	Portrait	4:5	--ar 4:5
Instagram Stories/Reels	Vertical	9:16	--ar 9:16
LinkedIn Post	Horizontal	1.91:1	--ar 191:100
YouTube Thumbnail	Horizontal	16:9	--ar 16:9
Twitter/X Post	Horizontal	16:9	--ar 16:9
Telegram Post	Horizontal	2:1	--ar 2:1
Pinterest	Vertical	2:3	--ar 2:3
OG Image (Web)	Horizontal	1.91:1	--ar 191:100

Output Template

### AI Image Prompts — [Content Name]

**Visual brief:** [link / brief description from $visual-brief]
**Tool:** [Midjourney / DALL-E 3 / Stable Diffusion / Flux]
**Platform:** [platform]
**Size:** [ratio]
**Brand palette:** [colors from $brand-guidelines — descriptively]

---

#### Prompt Anatomy

| Block | Value |
|------|----------|
| Subject | [specific description] |
| Action/Pose | [what they do] |
| Environment | [where, background] |
| Style | [from library] |
| Lighting | [type of light] |
| Palette | [brand colors description] |
| Composition | [angle, technique] |
| Mood | [emotion, atmosphere] |

---

#### Variant A — Primary (exactly to brief)

**Midjourney:**
\```
[full prompt] --ar [ratio] --v 6.1 --q 2 --s [value]
\```

**Negative:** `--no [exclusions]`

**DALL-E 3:**
"[descriptive prompt in natural language]"

---

#### Variant B — Alternative Style

**Midjourney:**
\```
[full prompt with different style] --ar [ratio] --v 6.1 --q 2
\```

---

#### Variant C — Alternative Composition

**Midjourney:**
\```
[full prompt with different angle] --ar [ratio] --v 6.1 --q 2
\```

---

#### Workarounds (if needed)
| Potential Issue | Applied Workaround |
|-----------------------|----------------------|
| [hands / text / faces] | [what was added to the prompt] |

#### Iteration Tips
- If too literal → add "artistic interpretation, conceptual"
- If wrong style → clarify: "in the style of [style]", add 2-3 style tags
- If oversaturated → add "muted tones, desaturated" + negative `--no oversaturated`
- If chaotic → add "minimalist, simple composition, negative space"

**→ Next step:** generation → review against `$content-review-checklist` (#25) → `$content-release-gate`