All skills
simota avatar

/sketch

@16a3f28
by shingo imotasimota/agent-skills85 stars
15

Generating AI image-generation code using the Gemini API. Handles text-to-image generation, image editing, and prompt optimization. Use when image generation code is needed.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/sketch

This session only. Nothing lands on disk.

referenceprompt-patterns.md

≈1.7k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Prompt Patterns Reference

Purpose: Read this when Sketch needs canonical prompt architecture, prompt-quality thresholds, JP -> EN translation rules, policy-safe rewriting, or v1.50+ prompt controls.

Contents

  • Prompt architecture
  • Style preset library
  • Domain templates
  • JP -> EN translation rules
  • Policy-safe prompting
  • Quality checklist
  • Negative patterns
  • SDK v1.50+ controls

Prompt Architecture

Build prompts in this order:

[Subject] + [Style] + [Composition] + [Technical]
Layer Role Weight Rule
Subject what to depict 40% put it first
Style mood and expression 30% keep style internally consistent
Composition framing and camera 20% choose ratio and perspective deliberately
Technical quality directives 10% keep this last and concise

Construction rules:

  1. subject first
  2. style immediately after the subject
  3. composition before technical polish
  4. technical keywords last

Style Preset Library

Photorealistic

photorealistic, natural lighting, detailed textures, DSLR quality, sharp focus, high dynamic range

Digital Illustration

digital illustration, clean lines, vibrant colors, flat design, modern graphic style

3D Render

3D render, isometric view, clean geometry, soft shadows, ambient occlusion, PBR

Watercolor

watercolor painting, soft edges, translucent layers, paper texture, bleeding colors

Abstract

abstract composition, geometric shapes, bold color palette, minimalist design, conceptual art

Additional styles:

  • oil painting
  • anime/manga
  • vintage/retro
  • cyberpunk
  • art nouveau
  • minimalist

Domain Templates

Marketing / hero image

[Product/Scene], professional commercial photography, hero image composition, brand-appropriate lighting

Recommended ratios:

  • 16:9
  • 21:9

Product photography

[Product] on [surface/background], studio product photography, soft box lighting, clean minimal composition

Recommended ratios:

  • 1:1
  • 4:3

UI / UX assets

[UI element/icon/illustration], flat design, UI-ready, consistent line weight, transparent-ready

Recommended ratio:

  • 1:1

Documentation / technical

[Concept visualization], clean infographic style, explanatory illustration, clear visual hierarchy

Recommended ratios:

  • 16:9
  • 4:3

Social media

Platform Ratio
Instagram Post 1:1
Instagram Story 9:16
Twitter/X 16:9
LinkedIn 4:3
YouTube Thumbnail 16:9

JP -> EN Translation Rules

Rules:

  1. restore omitted subjects
  2. replace vague adjectives with concrete English keywords
  3. expand mood words into recognizable visual descriptors
  4. keep technical terms in English

Common mappings:

Japanese English keywords
かわいい cute, adorable, charming, kawaii-style
かっこいい cool, stylish, sleek, dynamic
おしゃれ fashionable, trendy, chic, sophisticated
あたたかい(雰囲気) warm, cozy, inviting, golden tones
クール cool tones, blue palette, modern, minimal
レトロ vintage, retro, nostalgic, film-like
ナチュラル natural, organic, earthy, authentic
シンプル minimalist, clean, simple, uncluttered
ダイナミック dynamic, energetic, motion blur, bold angles
幻想的 ethereal, dreamy, fantasy, soft glow
荘厳 majestic, grand, imposing, awe-inspiring
繊細 delicate, intricate, fine detail, subtle

Keep this must-keep example:

Input (日本語):

やわらかい光のなかで、木のテーブルの上にコーヒーとノートパソコンがある、おしゃれなワークスペース

Output (English prompt):

A stylish modern workspace with a laptop and coffee cup on a wooden table, soft diffused natural lighting, cozy atmosphere, shallow depth of field, warm tones, lifestyle photography, 8K detail

Policy-Safe Prompting

Safe patterns:

  • use generic people descriptions instead of named real people
  • use generic workplaces instead of branded offices
  • use descriptive traits instead of copyrighted character names

Fallback sequence when a prompt is risky:

  1. remove sensitive keywords
  2. replace specific references with generic descriptions
  3. reframe with positive keywords
  4. simplify the prompt
  5. suggest an alternative concept if still blocked

Prompt Quality Checklist

  • subject is specific
  • style keywords do not conflict
  • aspect ratio matches the use case
  • no copyrighted character or brand reference unless explicitly required
  • person-generation policy is checked
  • prompt is in English
  • technical keywords are present
  • prompt length stays in the 50-200 word range

Negative Patterns

Style conflicts

Bad prompt Problem Fix
photorealistic watercolor painting photorealism and watercolor conflict choose one direction
minimalist detailed ornamental minimalist and ornamental conflict simplify or decorate, not both
bright dark moody cheerful tonal conflict choose one lighting and mood direction

Prompt length

Length Result Rule
< 20 words generic, vague results aim for at least 50
50-200 words best control and quality preferred range
> 200 words blurred subject, lower quality cut by importance
> 500 words keywords start losing force reduce aggressively

Ineffective patterns

Pattern Why it fails Better approach
NOT a photo of... negative instructions are weak use positive framing
Don't include people negation may be ignored empty scene, no figures, landscape only
ultra mega super high quality redundant modifiers cancel out use 8K detail, sharp focus
like the Mona Lisa copyright risk and vagueness describe the actual traits
beautiful amazing stunning subjective and weak use concrete technical instructions

Quality-keyword rule:

  • choose 3-5 strong quality keywords, not a long stack

SDK v1.50+ Controls

If v1.50+ is available, use:

image_generation_config=types.ImageGenerationConfig(
    aspect_ratio="16:9",
    person_generation="DONT_ALLOW",
)

Available parameters:

Parameter Values Default
aspect_ratio "1:1", "3:2", "2:3", "16:9", "9:16", "21:9" model default
person_generation "DONT_ALLOW", "ALLOW_ADULT" "DONT_ALLOW"

Version guidance:

  • v1.38+: control ratio and people via the prompt
  • v1.50+: prefer parameters for ratio and person generation, and keep the prompt focused on subject and style

Source: SKILL.md on GitHub

2 warnings4mo5 checks · Risk SAFE
  • Gen Agent Trust Hub4mo

    This skill provides a secure and well-architected framework for generating Python code to interact with the Google Gemini image generation API. It adheres to security best practices by emphasizing environment-variable-based credential management, providing explicit .gitignore guidance, and incorporating comprehensive error handling for API and safety filter responses. The skill acts exclusively as a code generator, ensuring that no actual API calls or external network requests are executed within the agent context itself.

  • Socket4mo

    No alerts

  • Snyk4mo

    Risk: MEDIUM · 1 issue

  • Runlayer6mo

    2/4 files flagged

  • ZeroLeaks5mo

    1 finding · Score: 86/100

Signed by skilld at 16a3f28. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated last month

README badge

README badge for simota/agent-skills/sketch