360 Secure SkillHub
360 安全技能中心

Image Create

该 Skill 根据文本需求生成图像提示词，先收集主题、风格、比例和限制条件，再给出安全、创意、极简三种提示词变体，并针对光照、构图、排版、色彩和细节进行迭代。适用于海报、缩略图、产品样机、Logo、概念艺术和社媒视觉。未声明具体图像生成工具、API 或运行环境。

wallaby16
wallaby16
热度 65.50
最近更新 2026-05-11 17:35:38
分类：创作应用场景：图像生成 / 品牌素材 / 社交媒体

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add image-create
Metadata
name
image-create
description
Generate images from text prompts and iterate on style/composition for posters, thumbnails, product mockups, logos, concept art, and social visuals. Use when the user asks to create a new image, restyle an image idea, produce prompt variants, or refine outputs with constraints like aspect ratio, mood, color palette, and text placement.

Image Create

Quick workflow

  1. Confirm goal in one line (what image should achieve).
  2. Collect minimum constraints:
    • subject
    • style
    • aspect ratio / resolution
    • must-include elements
    • must-avoid elements
  3. Produce 3 prompt variants:
    • safe/default
    • bold/creative
    • minimal/clean
  4. Generate image(s) with the selected variant.
  5. Iterate with precise edits (lighting, framing, typography, colors, detail level).

Prompt template

Use this template and fill all placeholders:

[TYPE OF IMAGE], featuring [MAIN SUBJECT], in [STYLE],
composition: [CAMERA/FRAMING], lighting: [LIGHTING],
background: [BACKGROUND], color palette: [COLORS],
mood: [MOOD], quality: high detail, clean edges,
output: [ASPECT RATIO OR SIZE].

Avoid: [NEGATIVE CONSTRAINTS]

Fast defaults

  • If user gives little detail, default to:
    • style: photorealistic (real camera look, natural skin texture, true-to-life materials)
    • lighting: soft natural light
    • composition: centered medium shot
    • lens/camera feel: 50mm, shallow depth of field, realistic dynamic range
    • avoid: CGI/plastic look, over-smoothing, blur, watermark, extra limbs, unreadable text, distorted faces
  • If text appears in image, keep wording short and ask for exact spelling.

Realistic style tuning

Use these additions whenever the user asks for realistic output:

  • Add quality cues: photorealistic, natural textures, physically plausible lighting, realistic shadows.
  • Add camera cues: shot on full-frame camera, 50mm lens (or 85mm for portraits).
  • Keep color grading subtle; avoid neon oversaturation unless requested.
  • Prefer environmental detail that supports realism (imperfections, material grain, natural reflections).

Iteration rules

  • Change one major variable per iteration when debugging quality.
  • Keep successful parts of previous prompt; only patch weak parts.
  • For brand visuals, preserve palette and typography constraints across iterations.
  • For thumbnails/posters, prioritize readability and high contrast.

Output format

When helping with generation requests, return:

  1. Primary prompt
  2. 2 alternatives
  3. Negative prompt
  4. Suggested settings (aspect ratio, style strength, number of outputs)
  5. One-line revision plan for next iteration