AI Image Prompts · 绘画提示词库

# ===== USER CONTROLS =====
title_text: "{argument name="display string" default="<<YOUR_TEXT>>"}"
title_style_preset: "{argument name="logo design preset" default="cool"}"  # cool, cute, cool_cute, luxury, pop, cyber, minimal, retro, kawaii
face_zoom: "{argument name="zoom on/off" default="off"}"            # off, on

subject: >
  Create a 21:6 wide banner using the provided reference image content.
  The primary face must be centered inside an invisible safe zone (not drawn),
  and a title logo (derived from title_text + title_style_preset) must be placed inside the same safe zone.

composition: >
  Aspect ratio 21:6.
  Invisible safe zone (DO NOT draw it):
  - Left margin: 15% of width
  - Right margin: 15% of width
  - Top margin: 16% of height
  - Bottom margin: 16% of height
  Keep the entire face AND the entire title logo fully inside this safe zone.
  Background may extend to the edges.

action: >
  Place the reference image content with maximum fidelity.
  Auto-crop/scale/reposition to fit a 21:6 banner with no stretching.
  Detect the primary face (if multiple faces, pick the largest/most prominent).
  Center the primary face within the safe zone both horizontally and vertically.
  Never crop off the face/eyes; keep the full face comfortably inside the safe zone (not near edges).

  Face zoom behavior:
  - If face_zoom is OFF: keep the original framing as much as possible while still centering the face in the safe zone.
  - If face_zoom is ON: zoom/crop to a bust-up portrait centered on the face
    (head and shoulders visible; include upper chest; do no

# ===== USER CONTROLS =====
title_text: "{argument name="display string" default="<<YOUR_TEXT>>"}"
title_style_preset: "{argument name="logo design preset" default="cool"}"  # cool, cute, cool_cute, luxury, pop, cyber, minimal, retro, kawaii
face_zoom: "{argument name="zoom on/off" default="off"}"            # off, on

subject: >
  Create a 3:1 wide banner using the provided reference image content.
  The primary face must be centered inside an invisible safe zone (not drawn),
  and a title logo (derived from title_text + title_style_preset) must be placed inside the same safe zone.

composition: >
  Aspect ratio 3:1.
  Invisible safe zone (DO NOT draw it):
  - Left margin: 15% of width
  - Right margin: 15% of width
  - Top margin: 16% of height
  - Bottom margin: 16% of height
  Keep the entire face AND the entire title logo fully inside this safe zone.
  Background may extend to the edges.

action: >
  Place the reference image content with maximum fidelity.
  Auto-crop/scale/reposition to fit a 3:1 banner with no stretching.
  Detect the primary face (if multiple faces, pick the largest/most prominent).
  Center the primary face within the safe zone both horizontally and vertically.
  Never crop off the face/eyes; keep the full face comfortably inside the safe zone (not near edges).

  Face zoom behavior:
  - If face_zoom is OFF: keep the original framing as much as possible while still centering the face in the safe zone.
  - If face_zoom is ON: zoom/crop to a bust-up portrait centered on the face
    (head and shoulders visible; include upper chest; do not crop the top of the head or chin; keep facial features prominent).

  Create negative space inside the safe zone for the title logo by adjusting crop/reposition/out

Read the motif depicted in the attached image, generate multi-angle shots, and create four scenes divided into a 2x2 grid.

[Action 1] The senior colleague (senpai) is looking at books on a shelf. The photographer presses a button on a remote control in their hand, and a distinctive electronic sound/chime rings. [Dialogue 1] The senpai is surprised, immediately turns to the camera, and playfully says, "Stop it." [Action 2] The photographer rings the electronic sound again. [Action 3] The senpai looks down shyly, squats momentarily, then stands up, laughing and covering their face with their hands. [Text Overlay] "{argument name="text overlay 1" default="Tried pressing the senpai's switch while they were shopping at the bookstore"}" "{argument name="text overlay 2" default="#This feeling is too cute www"}" "{argument name="text overlay 3" default="#Definitely inviting me www"}"

Add lots of Korean handwritten doodles, underlines, highlighters, and speech bubbles using a blue colored pencil. The doodles should primarily be compliments about the subject, and it's okay to add a few hand-drawn clip arts.

Create an image like the opening of Blade Runner

{
  "image_settings": {
    "aspect_ratio": "3:4",
    "resolution": {
      "width": 1152,
      "height": 1536
    }
  },

  "prompt": {
    "identity_lock": {
      "reference": "{argument name="reference" default="input_photo"}",
      "preserve": [
        "{argument name="preserve 1" default="face"}",
        "{argument name="preserve 2" default="facial proportions"}",
        "{argument name="preserve 3" default="eye shape"}",
        "{argument name="preserve 4" default="nose"}",
        "{argument name="preserve 5" default="lips"}",
        "{argument name="preserve 6" default="skin tone"}",
        "{argument name="preserve 7" default="skin texture"}",
        "{argument name="preserve 8" default="hairline"}",
        "{argument name="preserve 9" default="overall identity"}"
      ],
      "rules": [
        "{argument name="rule 1" default="no face swap"}",
        "{argument name="rule 2" default="no beautify"}",
        "{argument name="rule 3" default="no smoothing"}",
        "{argument name="rule 4" default="no reshaping"}",
        "{argument name="rule 5" default="no AI look"}"
      ]
    },

    "scene": {
      "camera_angle": "{argument name="camera angle" default="high-angle downward shot, POV"}",
      "composition": "{argument name="composition" default="MacBook screen fills most of the frame, thin strip of physical keyboard visible at the bottom"}",
      "screen_surface": [
        "{argument name="screen surface 1" default="visible RGB pixel grid"}",
        "{argument name="screen surface 2" default="subtle moire effect"}",
        "{argument name="screen surface 3" default="micro dust on glass"}",
        "{argument name="screen surface 4" default="faint fingerprints"}",
        "{argument name="screen surface 5" default="soft ambient reflections"}"
      ]
    },

    "digital_interface": {
      "os": "{argument name="os" default="macOS dark mode"}",
      "background_app": {
        "name": "{argument name="background app name" default="Spotify"}",
        "view": "{argument name="background app view" default="Liked Songs"}",
        "visible_tracks": [
          "{argument name="track 1" default="Blank Space – Taylor Swift"}",
          "{argument name="track 2" default="Shake It Off – Taylor Swift"}",
          "{argument name="track 3" default="Cruel Summer – Taylor Swift"}",
          "{argument name="track 4" default="Love Story – Taylor Swift"}"
        ]
      },
      "foreground_app": {
        "name": "{argument name="foreground app name" default="Photo Booth"}",
        "state": "{argument name="foreground app state" default="live preview window"}",
        "position": "{argument name="foreground app position" default="floating, center-right"}"
      }
    },

    "photo_booth_content": {
      "environment": {
        "room": "{argument name="room" default="dim bedroom"}",
        "background": "{argument name="background" default="off-white wall, rumpled bedding"}",
        "lighting": "{argument name="lighting" default="low-light, nocturnal, cool screen glow mixed with warm skin tones"}"
      },
      "subject": {
        "pose": "{argument name="subject pose" default="lying down, relaxed, candid"}",
        "expression": "{argument name="subject expression" default="natural, slightly relaxed"}",
        "outfit": "{argument name="subject outfit" default="light-colored tank top"}",
        "prop": {
          "item": "{argument name="prop item" default="iPhone 15 Pro"}",
          "hand": "{argument name="prop hand" default="right hand"}"
        }
      }
    },

    "realism_rules": [
      "{argument name="realism rule 1" default="this is a photo of a screen, not a screenshot"}",
      "{argument name="realism rule 2" default="raw smartphone photo look"}",
      "{argument name="realism rule 3" default="natural noise"}",
      "{argument name="realism rule 4" default="imperfect glass"}",
      "{argument name="realism rule 5" default="no studio lighting"}",
      "{argument name="realism rule 6" default="no HD polish"}"
    ]
  },

  "negative_prompt": [
    "{argument name="negative 1" default="screenshot"}",
    "{argument name="negative 2" default="flat UI"}",
    "{argument name="negative 3" default="perfect screen"}",
    "{argument name="negative 4" default="clean glass"}",
    "{argument name="negative 5" default="studio lighting"}",
    "{argument name="negative 6" default="beauty filter"}",
    "{argument name="negative 7" default="cartoon"}",
    "{argument name="negative 8" default="3d render"}",
    "{argument name="negative 9" default="painting"}",
    "{argument name="negative 10" default="watermark"}",
    "{argument name="negative 11" default="blurred face"}"
  ]
}

{  
 "image_layout": "2x2 grid collage featuring four distinct photographs of the same female subject.",   "subject_general": 
{     
"gender": "Female",     "hair_color": "Dark brown/black",     "hair_style": "Long, styled in loose waves in three panels; sleek high ponytail in one panel",     "aesthetic": "Glamorous, influencer, trendy, luxury lifestyle"   },   "panels": [     
{       
"position": "top_left",       "setting": "Outdoors at night, dark background with illuminated green foliage",       "action": "Subject is holding a cake with one hand and licking frosting off the index finger of the other hand",       "outfit": 
{        
 "top": "Black halter-neck sleeveless top",         "jewelry": "Gold wristwatch, gold bangle bracelet, ring, small hoop earrings"       },       "props": 
{         
"item": "Round white frosted cake",         "details": [           "Red cherries on top",           "Decorative white piping along edges",           "Black icing text reading 'bad bitch energy'"         ]       }     },     
{       
"position": "top_right",       "setting": "Daytime city street, likely an upscale shopping district (e.g., Rodeo Drive) with palm trees and storefronts visible",       "action": "Subject is leaning out of the open door of a black luxury car, looking at the camera",       "outfit": 
{         
"top": "White sleeveless ribbed crop top",         "bottom": "Blue denim jeans",         "accessories": "Shoulder bag (strap visible), gold hoop earrings"       },       "lighting": "Bright natural sunlight casting shadows"     },     
{      
 "position": "bottom_left",       "setting": "Indoor hallway, beige walls, wooden floor",       "action": "Full-body mirror selfie",       "outfit": 
{         
"style": "Matching two-piece set",         "color": "Dark brown/chocolate",         "top": "Velvet/suede corset-style bustier top",         "bottom": "Mini skirt",         "accessories": "Patterned designer handbag (resembling Dior Saddle bag), stack of gold bracelets"       },       "background_details": [         "Large rectangular mirror with gold frame leaning against wall",         "Vase with pampas grass in the corner",         "Doorway visible in reflection"       ]     },     
{       
"position": "bottom_right",       "setting": "Inside a vehicle or outdoors, golden hour sunlight",       "action": "Close-up selfie, hand raised near mouth/chin",       "outfit": 
{         
"top": "Light pink/mauve ribbed tank top",         "outerwear": "Beige fuzzy/sherpa jacket worn off-the-shoulder",         "jewelry": "Layered gold necklaces with pendants, hoop earrings"       },       "hair_styling": "Sleeked back high ponytail",       "background_details": "Car window frame, palm trees and blue sky visible in background"     
}  
 ]
  }

Please illustrate the features of ChatGPT 5.2 using perfect Japanese fonts. The ratio should be 16:9. Please also add cute illustrations.

Actually, there's a trickier way to use this.

First, have it systematically summarize the information
→ Then, based on that, have it create the output/deepen the discussion
Just doing this makes the quality of the output completely different.

This is because if you first create a blueprint in the AI's mind—including "terminology," "judgment criteria," "prohibited items," "patterns," and "priorities"—

The next instructions will be output according to that blueprint, making it less likely to deviate.

For example, if you want Nano Banana Pro to create a thumbnail:

① First, ask GPT-5.2:
"Thoroughly research and systematize the best practices for creating prompts for YouTube thumbnails using Nano Banana Pro. Include common mistakes as well."

② After that, ask:
"Based on the current system, create a prompt for making a YouTube thumbnail on the theme of {argument name="theme" default="XX"} using Nano Banana Pro. The video content is like this, and I want to include {argument name="wording" default="this kind of wording"}."

Then, input the completed prompt into Nano Banana Pro. It becomes much easier to get the desired output.

If you just ask it to "create" every time, the AI outputs based on the moment, leading to inconsistent results. But if you create a map (system) first, it's powerful.

Moreover, this method works for side business strategies, sales copy, and video scripts. The versatility is insane.

I highly recommend having the AI organize how to create good output before having it perform the task, and then having it think based on that organization.

By the way, the thumbnail created this time is the third image.
Since the first attempt was mediocre, I had it systematize how to create attractive YouTube thumbnails before generating the prompt.

layering:
  order:
    - text_layer
    - character_layer
    - background_layer
  strict: true

subject:
  description: >
    Preserve the subject exactly as in the reference image, keeping all facial features,
    colors, hairstyle, clothing, and proportions. No redesign or reinterpretation.
  framing: >
    Bust-up composition (head to upper torso), making the subject appear large and prominent.
  integrity: full_identity_preservation
  detail_level: ultra_high_resolution
  micro_details: preserve_skin_texture_hair_edges_fabric_details

composition:
  aspect_ratio: "16:9"
  layout: >
    Subject positioned on one side; headline text on the opposite side.
  emphasis: >
    Strong presence on the character's upper body and expression.

action:
  description: >
    Minimal motion enhancement. Maintain the reference pose while adding subtle dramatic energy.

location:
  description: >
    Simple background created by interpreting the reference image’s mood.
    Use soft gradients, atmospheric glow, light particles, or abstract light rays.
    Avoid complex textures or any copyrighted shapes.
    Background must NOT contain any characters, silhouettes, faces, bodies,
    or blurred forms derived from the subject.
  background_feel: >
    Subtle, atmospheric, clean.
  allowed_elements: pure_color_gradients_light_abstract_shapes_only
  prohibit_subject_reuse_in_background: true

style:
  visual_style: >
    High-impact reveal-splash aesthetic with strong contrast between character and headline.
  text_effects: >
    Metallic, high-energy lettering with pronounced depth and perspective. No glow effects.
  character_rendering: >
    Extremely high-resolution rendering with sharp facial details, clean hair definition,
    precise edges, and crisp texture fidelity.
  background_style_rules: >
    Background must be purely non-figurative. No silhouettes, no blurred shapes,
    and no reused subject elements.
  ink_splash_rules: >
    Ink splash must be pure black only. No other colors, tinting, blending, or variation.

camera_lighting:
  camera_angle: from_reference
  lighting: >
    Enhance with dramatic key lighting and subtle rim highlights to emphasize the reveal.
  clarity: maximum_definition

colors:
  palette: >
    Background colors inferred from the reference image.
    Character name in metallic silver; announcement text in saturated yellow.
  mood: dramatic, heroic, high_energy

text:
  character_name: "{argument name="character name" default="キャラクター名"}"
  announcement_fixed: "Participate!!"
  placement: opposite_side_of_character
  p

You are an expert in the "Adaptive Salience Model," creatively integrating knowledge from cognitive science (neuroscience, psychology, behavioral economics), and you are a top-tier art director.

Deeply analyze the [Input Text (Article Title or Beginning)] provided at the end, and generate a thumbnail image that maximizes the CTR (Click-Through Rate) for that article.

If a reference image is attached, place that character as the "person presenting the article content" and maintain consistency.

▼ Instructions for Generation Process (Thinking Process)
Use Nano Banana Pro's "Thinking Mode" to logically formulate the composition plan in the following steps before drawing:

1.  **[Step 1: Content Analysis and Hook Extraction]**
    * Identify the "maximum benefit the reader gains" or the "risk of loss if they don't know" from the input text (Application of Prospect Theory).
    * Select the most intuitive visual metaphor symbolizing the article's theme (e.g., rise, key, lightning, light, darkness, etc.).

2.  **[Step 2: Optimization by Adaptive Salience Model]**
    * **FFA Activation (Face Fusiform Area)**: Maximize the resolution of the character's face, especially the "eyes," and design the composition so that the gaze meets the viewer's (eye contact). Match the expression to the article's tone (surprise, confidence, seriousness, etc.).
    * **Pop-out Effect**: Use "vivid colors (neon colors or yellow, etc.)" that are complementary to the background color for the title text and important icons to create visual prominence.
    * **Cognitive Load Theory**: Do not overload the information. Limit elements to "1 face + 1 impactful text + 1 background metaphor."

3.  **[Step 3: Layout Design]**
    * Design the layout, considering the Z-pattern or F-pattern, so that the viewer's gaze flows from the "face" → "high jump-rate text" → "clickable element."

▼ Design and Style Specification
* **Composition**: Close-up (bust shot) or medium shot. Place the character in the center or at the intersection of the rule of thirds.
* **Taste/Aesthetic**: [Optimize for the genre of the input text. Examples: For business, "sophisticated, trustworthy"; for entertainment, "vibrant, high saturation"; for technology, "cyberpunk, futuristic"]
* **Text Rendering (Important)**:
    * Extract the **most catchy short phrase (recommended 10 characters or less)** from the input text.
    * **Font**: Use "**Heavy Serif**" font. To enhance readability, the text must have a **thick outline** contrasting with the background color.
    * **Placement (Jump Rate)**: **Set a high "Jump Rate."** Place the main keywords extremely large, and auxiliary words or sub-information small, creating visual rhythm and contrast.
* **Character Direction**: If there is a reference character, adjust their attire to match the article's theme (lab coat, suit, explorer, casual, etc.).

▼ Input Text
{argument name="input text" default="(Article title or content)"}

Astronaut floating in space, bold title text '{argument name="title text" default="BEYOND THE VOID"}' layered between character and nebula background, cinematic lighting, dramatic composition