AI 资产台账

定义：一张记录每份资产来源与授权的表格。

Definition: a table recording the source and licence of every asset.

为什么它比你想的更重要

⚠ AI 生成资产的商用授权通常不明确 —— 模型训练用了什么数据、

输出能不能商用，取决于服务条款，而条款随时会变。

半年后你想上架时，才发现当初用的那个服务改了条款 —— 你无从得知。

Why it matters more than you would think

⚠ The commercial rights of AI-generated assets are usually unclear — what data the model was trained on, and whether you may sell its output, comes down to the service terms. And terms change.

Six months later, when you try to ship, you find out the service you used changed its terms — and you had no way of knowing.

最简格式

规则：发布前「待确认」列必须清零。

The simplest useful format

Rule: before you ship, the "unconfirmed" column must be empty.

三条实操规则

1. 程序化生成最安全

代码画出来的图形版权归你，没有第三方来源问题。

程序化美术（procedural art）在这一层优势明显。

2. ⚠ 字体最常被忽略

每个人都会检查图片和音频的授权，很少有人检查字体。

而绝大多数免费字体的授权是「个人免费」，不含商用。

3. CC-BY 需要署名

用了 CC-BY 素材而没署名，可能违反协议。

混用时注意保留署名信息。

Three practical rules

1. Procedural generation is the safest

Code-drawn shapes are yours; there is no third-party provenance question. Procedural art has a clear advantage here.

2. ⚠ Fonts are the most forgotten

Everyone checks the licence on images and audio. Almost nobody checks fonts — and most "free" fonts are free for personal use, not for commercial use.

3. CC-BY requires attribution

Using CC-BY material without crediting it may breach the licence. Keep the attribution when you mix.

⚠ 一个常见误解

「AI 生成的就都是我的」

⚠ 不一定。模型训练数据的来源通常不明确，

「输出属于用户」是服务方给的授权，不是版权法的结论。

⚠ A common misconception

"AI made it, so it's mine"

⚠ Not necessarily. The provenance of the training data is usually unclear, and "the output belongs to you" is a grant from the service provider, not a conclusion of copyright law.

什么时候建

从第一天就建。 事后补是补不回来的 —— 你已经想不起来

三周前那张图是从哪拿的了。

When to start it

Day one. You cannot backfill it later — by then you will not remember where the image from three weeks ago came from.