WatermarkConfig
Options & Configuration · AiDotNet.Safety
Configuration for watermarking modules.
For Beginners
Watermarking embeds invisible markers in AI-generated content so it can later be identified as AI-generated. This is increasingly required by regulations like the EU AI Act (Article 50).
How It Works
References:
- SynthID-Text: Production text watermarking at scale (Google DeepMind, Nature 2024)
- SynthID-Image: Internet-scale image watermarking (Google DeepMind, 2025)
- Only 38% of AI generators implement adequate watermarking (2025)
Constructors
| Constructor | Summary |
|---|---|
WatermarkConfig() |
Properties
| Property | Summary |
|---|---|
IGenerationTokenizer TextTokenizer { get; set; } |
Gets or sets the tokenizer of the language model whose output carries the generation-time text watermark. |
bool? AudioWatermarking { get; set; } |
Gets or sets whether audio watermarking is enabled. |
bool? DetectionMode { get; set; } |
Gets or sets watermark detection mode (detect existing watermarks on input). |
bool? ImageWatermarking { get; set; } |
Gets or sets whether image watermarking is enabled. |
bool? TextWatermarking { get; set; } |
Gets or sets whether text watermarking is enabled. |
double? WatermarkStrength { get; set; } |
Gets or sets the watermark strength, in (0, 1]. |
string ImageWatermarkKey { get; set; } |
Gets or sets the secret key of the image watermark: only a watermarker with the same key embeds or detects it. |
string TextKey { get; set; } |
Gets or sets the secret key of the generation-time text watermark. |