All-in-One GPT Image 2 Prompt Templates – From Basics to Advanced Reusable Skeleton

Golden Formula: 8 Elements

The core prompt structure consists of eight elements that operate behind the scenes for any image generation request.

Format – aspect ratio and orientation (e.g., 16:9, 9:16, 1:1)

Subject – concrete noun describing the main object

Setting – environment, background, foreground (location + desk + weather + props)

Style – visual or artistic style

Lighting – type and direction (e.g., golden hour, studio lighting, soft side light)

Lens/Composition – perspective, focal length, depth of field (e.g., wide angle, macro, shallow depth of field)

Text/Negative – what to add or exclude (e.g., NO people, NO watermark)

Use Case – intended purpose and size (e.g., WeChat cover, e‑commerce hero, social media post)

prompt = (
    "[Format] [Subject], [Setting], "
    "[Style], [Lighting], [Lens/Composition], "
    "[Text constraints], [Color constraints], "
    "NO [Undesired element 1], NO [Undesired element 2]"
)

# Example: WeChat cover
cover_prompt = (
    "9:16 vertical, a young entrepreneur sitting at a desk, "
    "city skyline outside the window, warm golden hour light, "
    "cinematic, golden hour backlight, wide angle lens, "
    "centered composition, navy and gold color scheme, "
    "NO cluttered background, NO watermark, NO text"
)

Tip: Treat the eight elements as a mental checklist; filling more of them generally yields more reliable outputs.

Style Cheat Sheet (15 Common Styles)

Key style descriptors with ready‑to‑copy prompt fragments.

Cinematic :

cinematic, 16:9 widescreen, dramatic lighting, film color grading

Product Photography :

product photography, pure white background, studio lighting, sharp focus

Food Photography :

food photography, 45-degree overhead, warm tones, shallow depth of field

Flat Illustration :

flat illustration, solid colors, no gradients, minimal details

Sketch Whiteboard :

sketch whiteboard style, blue hand‑drawn lines, orange annotations, light yellow background

Minimalist :

minimalist, negative space, geometric, monochrome with one accent color

Cyberpunk :

cyberpunk, neon lighting, rain‑slicked streets, high contrast, purple and cyan

Combination rule: Avoid contradictory pairs such as cyberpunk + watercolor or flat illustration + photorealistic . Recommended combos include isometric illustration + pastel colors + clean and cinematic + golden hour + widescreen .

Scene Cheat Sheet (7 High‑Frequency Templates)

Ready‑to‑use prompts for the most common production needs.

# 1. WeChat cover (9:16 vertical)
prompt = "poster design, minimalist, centered composition, negative space on sides, navy and gold, Morandi tones, soft studio lighting, title area reserved at top, NO watermark"
# size = "1024x1536"

# 2. Article illustration (16:9 horizontal)
prompt = "wide landscape, golden hour, warm tones, shallow depth of field, rule of thirds, cinematic, anamorphic lens feel, NO people, NO text"
# size = "1536x1024"

# 3. E‑commerce product (1:1 square)
prompt = "product photography, pure white background, studio lighting, centered composition, 45-degree angle, sharp focus, NO shadows, NO reflections"
# size = "1024x1024"

# 4. Infographic (vertical)
prompt = "infographic layout, flat vector illustration, mint green and white, clean minimal, three‑column card layout, centered title area, NO photorealistic"
# size = "1024x1536"

# 5. Social media post (square)
prompt = "square format, soft pastel, flat illustration, centered composition, duotone, negative space for text, NO photorealistic, NO busy background"
# size = "1024x1024"

# 6. Tech‑style illustration (horizontal)
prompt = "futuristic tech scene, cyberpunk, neon lighting, cool tones with cyan and purple, low angle, wide lens, high contrast, NO text, NO logos"
# size = "1536x1024"

# 7. Portrait/headshot
prompt = "portrait photography, professional headshot, Rembrandt lighting from upper left, shallow depth of field, warm skin tones, 85mm lens, clean blurred background"
# size = "1024x1024"

JSON Structured Template

A seven‑field JSON skeleton standardises prompt data for batch generation and ensures consistent style across outputs.

{
  "subject": "主体描述",
  "style": "视觉风格",
  "lighting": "光线类型",
  "composition": "构图方式",
  "colors": "色彩方案",
  "mood": "氛围情绪",
  "output": "输出规格"
}

Typical values for each field include: style:

product photography

lighting

:

studio diffused light

colors

:

white + 1 accent

mood

:

premium, clean

output

:

1:1 e‑commerce hero image

BASE_TEMPLATE = {
    "lighting": "soft studio diffused light",
    "composition": "centered, 45-degree overhead",
    "colors": "white, navy blue accent",
    "mood": "premium, professional",
    "output": "1:1 e‑commerce hero image"
}

products = [
    {"subject": "black wireless earbuds", "style": "product photography"},
    {"subject": "silver smartwatch", "style": "product photography"},
    {"subject": "brown leather wallet", "style": "product photography"}
]

for p in products:
    prompt_json = {**BASE_TEMPLATE, **p}
    # call API to generate image

Why JSON? Guarantees batch‑wide style consistency and lets a single parameter be tweaked without affecting the others.

Advanced Techniques Cheat Sheet

Multi‑round iteration : Change only one variable per round, lock all others, and optionally restart after four rounds to avoid degradation.

Local editing (3W1K formula) :

What to change – specific element

What attribute – colour, material, shape, position

What to keep – list of locked elements

Key constraint – stronger negative words

Example: change only the colour of the second vase from white ceramic to dark green glass while keeping background, lamp, window light, and overall tone unchanged.

Reference‑image usage : Three modes – style reference, character reference, object reference – each expressed as “use the uploaded image for style/face/appearance, keep other aspects constant”.

Multi‑image generation strategy :

Phase 1 (exploration): generate n=4 variants to gauge direction.

Phase 2 (confirmation): generate a single image and fine‑tune the prompt.

Phase 3 (character series): repeat the same character description for each image, generating one at a time.

Common Errors – 10‑Point Red‑Line Checklist

Vague description → Apply the 8‑element formula and write at least five elements.

Changing too many variables at once → Modify only one variable per iteration.

Ignoring aspect ratio → Specify size explicitly, e.g., size="1536x1024".

Stacking style words → Use one primary style plus up to two modifiers.

Prompt too long → Keep main description 80‑120 characters; place constraints in a NO … clause.

Missing lighting → Include at least one lighting type and direction.

No negative constraints → Append NO X, NO Y to exclude unwanted elements.

Unintended side effects when editing → Use the 3W1K formula to list immutable items.

Inconsistent characters across images → Repeat full character description in each prompt.

No systematic checklist → After generation, run a seven‑step quality check: subject clarity, lighting consistency, composition, style uniformity, detail integrity, absence of unwanted elements, overall tone suitability.

Quality Check – 7‑Step Post‑Generation Checklist

Subject is clear and not deformed.

Lighting direction is consistent, no contradictory light sources.

Composition matches expectations (ratio, subject placement).

Style is uniform (no conflicting style mixes).

Details are correct (hands, text, logos).

No unwanted elements appear.

Overall colour tone matches the intended use case.

Universal Skeleton – End‑to‑End Decision Chain

Step‑by‑step process:

1️⃣ Define need – what is the image for? (cover, illustration, e‑commerce, infographic, avatar)
2️⃣ Choose format – 9:16 vertical / 16:9 horizontal / 1:1 square
3️⃣ Pick style – photography / illustration / art (select 1‑2 from the 15)
4️⃣ Set lighting – natural (6 options) or artificial (6 options)
5️⃣ Choose lens – wide / standard / macro / telephoto + perspective (high / eye‑level / low)
6️⃣ Decide colors – warm / cool / saturated / pastel / Morandi / black‑white + palette
7️⃣ Determine composition – centered / rule‑of‑thirds / poster / infographic / card
8️⃣ Add constraints – <code>NO X, NO Y, avoid Z</code>
9️⃣ Generate image and run the 7‑point checklist

Final one‑sentence prompt skeleton (text version):

[Format] [Subject] in/on [Setting], [Style], [Lighting], [Lens], [Colors], [Composition], NO [Exclude 1], NO [Exclude 2]

JSON version mirrors the same fields:

{
  "format": "16:9 horizontal",
  "subject": "主体描述",
  "setting": "场景描述",
  "style": "风格，辅助修饰1，辅助修饰2",
  "lighting": "光线类型 and 方向",
  "lens": "镜头 + 视角",
  "colors": "色彩方案",
  "composition": "构图方式",
  "negative": ["排除项1", "排除项2", "排除项3"],
  "output": "最终用途和尺寸"
}