asset-generation

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Asset Generation

资产生成

Use this skill for logos, icon marks, hero/section imagery, and texture motifs.

本技能适用于logo、图标标识、主视觉/板块图像以及纹理图案的生成。

Trigger

触发条件

Run when user asks for:

logo exploration or logo revisions
better visual assets in design proposals
proposal-specific imagery
provider comparison (Recraft, OpenAI, Gemini/Nano Banana)

当用户提出以下需求时运行：

logo探索或logo修订
设计提案中更优质的视觉资产
提案专属的图像素材
生成工具对比（Recraft、OpenAI、Gemini/Nano Banana）

Core Rules

核心规则

Use all three providers during exploratory rounds unless user opts out:
- Recraft
- OpenAI Image (
```
gpt-image-1
```
  )
- Nano Banana Pro (Gemini image model)
Generate assets in proposal context, not as detached global batch.
Exploration width must match design uncertainty:
- Early rounds: high style/color diversity
- Late rounds: narrow around survivors
Logos must pass favicon/app-icon constraints:
- flat, simple, low-part count
- legible at 16px/24px
- works one-color and full-color
If provider outputs are noisy/illustrative, use a deterministic geometric SVG fallback.
Treat user-provided logo references as quality primitives only, never literal templates.
Never produce marks that can be confused with existing company logos.

除非用户明确要求，否则在探索阶段必须同时使用三款生成工具：
- Recraft
- OpenAI Image (
```
gpt-image-1
```
  )
- Nano Banana Pro（Gemini图像模型）
生成的资产必须贴合提案上下文，而非独立的批量生成内容。
探索广度需与设计不确定性匹配：
- 早期阶段：高风格/色彩多样性
- 后期阶段：聚焦于筛选出的优质方案
Logo必须满足网站图标/应用图标约束：
- 扁平化、简洁、元素数量少
- 在16px/24px尺寸下仍清晰可辨
- 支持单色和全色彩显示
若工具输出内容偏噪点/插画风格，需使用确定性几何SVG作为备选方案。
用户提供的logo参考仅作为质量基准，绝不能作为字面模板直接使用。
禁止生成可能与现有企业logo混淆的标识。

Reference Archetypes (Quality North Star)

参考原型（质量标杆）

Target mark families like:

modular rounded blocks
orbiting dot constellation
constrained grid/mosaic tiles
minimalist key/keyhole symbol
3x3 dot matrix
hex-aperture emblem

Avoid mascot illustrations, scene composition, and decorative line clutter.

目标标识风格包括：

模块化圆角块
环绕式点群
受限网格/马赛克 tiles
极简钥匙/锁孔符号
3x3点阵
六边形开口徽章

避免吉祥物插画、场景构图和装饰性线条冗余。

Reference Use Policy (Hard)

参考使用规则（严格执行）

When user provides inspiration logos:

Extract only abstract qualities:
- simplicity level
- geometry type
- corner behavior
- stroke/fill balance
- negative-space strategy
Do not reuse exact silhouette, arrangement, or recognizable motif from references.
Do not preserve source colorways unless the user explicitly asks.
Generate at least 4 structural families per batch before narrowing.
Reject anything that looks like "same logo with minor tweaks."

This is a hard requirement. "Inspired by" means transferable principles, not copied form.

当用户提供灵感logo时：

仅提取抽象特质：
- 简洁程度
- 几何类型
- 边角处理方式
- 描边/填充平衡
- 负空间策略
禁止复用参考素材的精确轮廓、布局或可识别的图案。
禁止保留参考素材的配色方案，除非用户明确要求。
每批生成至少4种结构风格后再进行筛选。
拒绝任何看起来像是“原logo微调版”的方案。

这是硬性要求。“灵感来源于”指的是可迁移的设计原则，而非复制形态。

Complexity Budget (Hard Limits)

复杂度预算（严格限制）

Reject marks that exceed:

8 primary primitives
2 stroke widths
2 brand colors + 1 neutral
more than 1 internal detail per major shape

Accept only marks that remain recognizable at 24px without anti-aliasing blur.

拒绝超出以下限制的标识：

超过8个基础几何元素
超过2种描边宽度
超过2种品牌色 + 1种中性色
每个主要形状内的细节超过1处

仅接受在24px尺寸下无抗锯齿模糊仍可识别的标识。

Phase-Aware Breadth

阶段感知的广度策略

Phase A: Wide Exploration (no clear winner yet)

阶段A：广度探索（未确定最优方案）

Per proposal, generate at least:

1 Recraft concept
1 OpenAI concept
1 Nano Banana concept

Diversity matrix (required across batch):

Geometry: rounded / angular / mixed
Weight: monoline / medium / bold
Symbol strategy: abstract / mic-derived / node-derived
Palette families: warm, cool, neutral, high-contrast, muted

针对每个提案，至少生成：

1个Recraft方案
1个OpenAI方案
1个Nano Banana方案

多样性矩阵（每批必须覆盖）：

几何风格：圆角/棱角/混合
视觉重量：单线条/中等/粗体
符号策略：抽象/麦克风衍生/节点衍生
调色板系列：暖色调、冷色调、中性色调、高对比度、低饱和度

Phase B: Convergence (2-3 survivors)

阶段B：收敛聚焦（保留2-3个优质方案）

Per proposal, generate:

2 focused variants (best provider + fallback provider)
tightly constrained palette and stroke system

针对每个提案，生成：

2个聚焦变体（最优工具 + 备选工具）
严格约束的调色板和描边系统

Provider Routing

工具路由

Task	Primary	Secondary
Minimal vector-like symbol mark	Recraft ( `icon` or simple `logo` )	OpenAI
Text-sensitive wordmark lockups	Nano Banana Pro	OpenAI
Stylized hero/section art	Nano Banana Pro	OpenAI/Recraft raster
Subtle background motifs	Recraft raster / Nano Banana	OpenAI

任务	首选工具	备选工具
极简类矢量符号标识	Recraft（ `icon` 或简单 `logo` 模式）	OpenAI
对文本敏感的文字标识组合	Nano Banana Pro	OpenAI
风格化主视觉/板块图像	Nano Banana Pro	OpenAI/Recraft 栅格图
微妙背景图案	Recraft 栅格图 / Nano Banana	OpenAI

Logo Prompt Contract

Logo提示词规范

Always include:

```
no text
```
```
no mockup
```
```
no shadows
```
```
no gradients
```
(unless explicitly testing gradient direction)
```
centered symbol
```
```
flat icon
```

Template:

[brand intent], [shape language], minimalist flat icon logo, no text, no mockup, no shadows, no gradients, centered symbol, transparent or plain background

Always include:

```
no text
```
```
no mockup
```
```
no shadows
```
```
no gradients
```
(unless explicitly testing gradient direction)
```
centered symbol
```
```
flat icon
```

模板:

[品牌定位], [形态语言], minimalist flat icon logo, no text, no mockup, no shadows, no gradients, centered symbol, transparent or plain background

Proposal-Context Workflow

提案上下文工作流

For each proposal

7a..7h

Read proposal DNA + palette + typography mood.
Generate 3 logos (Recraft/OpenAI/Nano Banana) using that proposal context.
Pick best mark for that proposal only.
Embed into that proposal HTML and tune surrounding spacing/contrast.
Keep alternates for review.

Do not force one logo family across all proposals during exploratory rounds.

For each proposal

7a..7h

读取提案核心信息 + 调色板 + 排版风格。
结合提案上下文生成3个logos（Recraft/OpenAI/Nano Banana各一个）。
仅为该提案挑选最优标识。
将其嵌入该提案HTML并调整周边间距/对比度。
保留备选方案供审核。

Do not force one logo family across all proposals during exploratory rounds.

QA Gates (Reject if fails)

noisy illustration masquerading as logo
too many micro-details for favicon scale
style clashes with proposal typography/system
same colorway/style duplicated across whole batch
visible gibberish text baked into background assets
mascot/character or scene illustration instead of symbol mark
more than one focal object in a logo frame
obvious resemblance to known logos or user-provided samples
structural near-duplicate across proposal logos

noisy illustration masquerading as logo
too many micro-details for favicon scale
style clashes with proposal typography/system
same colorway/style duplicated across whole batch
visible gibberish text baked into background assets
mascot/character or scene illustration instead of symbol mark
more than one focal object in a logo frame
obvious resemblance to known logos or user-provided samples
structural near-duplicate across proposal logos

Deterministic Fallback

When model outputs fail QA, generate a clean SVG mark from geometric primitives. This is not optional for quality: use it to guarantee a production-safe floor.

Output Layout

Use proposal-scoped paths:

.design-evolution/gen-N/assets/logos/7a/recraft-1.svg

.design-evolution/gen-N/assets/logos/7a/openai-1.png

.design-evolution/gen-N/assets/logos/7a/gemini-1.png

.design-evolution/gen-N/assets/textures/7a/*.png

Chosen production candidate:

.design-evolution/gen-N/assets/logos/7a/final.(svg|png)

Use proposal-scoped paths:

.design-evolution/gen-N/assets/logos/7a/recraft-1.svg

.design-evolution/gen-N/assets/logos/7a/openai-1.png

.design-evolution/gen-N/assets/logos/7a/gemini-1.png

.design-evolution/gen-N/assets/textures/7a/*.png

Chosen production candidate:

.design-evolution/gen-N/assets/logos/7a/final.(svg|png)

Minimal Command Patterns

Recraft (symbol mark)

bash

python3 "$ENGINE" --repo "$REPO" recraft icon \
  "minimal flat icon mark for Vox Cloud, abstract mic+node, no text, no gradients" \
  --colors "#1E5CFF,#111827" --n 3 --out "$OUT"

bash

python3 "$ENGINE" --repo "$REPO" recraft icon \
  "minimal flat icon mark for Vox Cloud, abstract mic+node, no text, no gradients" \
  --colors "#1E5CFF,#111827" --n 3 --out "$OUT"

OpenAI (

gpt-image-1

)

OpenAI (

gpt-image-1

)

bash

curl -sS https://api.openai.com/v1/images/generations \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-1","prompt":"minimal flat icon logo mark, no text, no gradients, no shadows","size":"1024x1024","background":"transparent"}'

bash

curl -sS https://api.openai.com/v1/images/generations \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-1","prompt":"minimal flat icon logo mark, no text, no gradients, no shadows","size":"1024x1024","background":"transparent"}'

Nano Banana Pro (Gemini)

bash

curl -sS "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent?key=$GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"minimal flat icon logo mark, no text, no gradients"}]}],"generationConfig":{"responseModalities":["TEXT","IMAGE"]}}'

bash

curl -sS "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent?key=$GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"minimal flat icon logo mark, no text, no gradients"}]}],"generationConfig":{"responseModalities":["TEXT","IMAGE"]}}'

asset-generation

Original

Translation

Asset Generation

资产生成

Trigger

触发条件

Core Rules

核心规则

Reference Archetypes (Quality North Star)

参考原型（质量标杆）

Reference Use Policy (Hard)

参考使用规则（严格执行）

Complexity Budget (Hard Limits)

复杂度预算（严格限制）

Phase-Aware Breadth

阶段感知的广度策略

Phase A: Wide Exploration (no clear winner yet)

阶段A：广度探索（未确定最优方案）

Phase B: Convergence (2-3 survivors)

阶段B：收敛聚焦（保留2-3个优质方案）

Provider Routing

工具路由

Logo Prompt Contract

Logo提示词规范

Proposal-Context Workflow

提案上下文工作流

QA Gates (Reject if fails)

QA Gates (Reject if fails)

Deterministic Fallback

Deterministic Fallback

Output Layout

Output Layout

Minimal Command Patterns

Minimal Command Patterns

Recraft (symbol mark)

Recraft (symbol mark)

OpenAI (gpt-image-1)

OpenAI (gpt-image-1)

Nano Banana Pro (Gemini)

Nano Banana Pro (Gemini)

OpenAI (
`gpt-image-1`
)

OpenAI (
`gpt-image-1`
)