Note that in the preview releases, configuring the server is an advanced process that assumes some familiarity with editing configuration files yourself. This process will hopefully improve in future releases so it is only recommended to proceed if you are comfortable with that, and really need to.

Configuration Files

The server will look for configuration in a handful of locations when it starts up, but it's recommended to use configuration files stored in the same directory as the server (i.e. the same folder where BetaCensor.Server is located).

You can use either JSON or YAML configuration files, which should be named config.json or config.yml respectively.

While we strongly recommend you manually build your own configuration file with only the parts you need, you can see an example of (almost) all the configuration options available to the server in this example configuration.

Sticker Configuration

You can enable stickers and control their use using the Stickers section of the config file. You can either use the config.yml/config.json or use a separate stickers.yml/stickers.json. You can configure two kinds of locations for stickers: sticker stores and separate paths.

While this might seem a little complex, it should be easy to get set up once and means you can use any of your own images as stickers, no matter where they're stored.

Stores

A "sticker store" is a folder with stickers for more than one category, with directories for each category available in the store. For example, a store might be laid out like this:

C:/path/to/my/sticker/store
└───Professional
    │  squareprocap1.png
    │  squareprocap2.png
    │  squareprocap4.png
└───Chastity
    │  chastity_sticker1.png
    │  ch_sticker2.png
    │  ch_sticker3.png

Default Store

Beta Censoring looks for a default store at the stickers/ path from the server folder. That is, if there's a folder called stickers in the same place as BetaCensor.Server.exe, Beta Censoring will automatically load any subfolders in that folder as categories of stickers.

Likewise, it will also load any ZIP files with the .betapkg extension or with names ending in -stickers.zip. That is, if you have a ZIP file named my-beta-stickers.zip in the same folder as BetaCensor.Server.exe, Beta Censoring will automatically load any folders in that ZIP file as categories of stickers.

Configuration

You can also add sticker stores to your configuration like below, and it would make a Professional and Chastity category available (using the sample layout from above).

Stickers:
  LocalStores: ['C:/path/to/my/sticker/store']

{
    "Stickers": {
        "LocalStores": ["C:/path/to/my/sticker/store"]
    }
}

If you have a Beta Safety installation handy, you can always add its browser-extension/images/stickers path as a store to Beta Censoring.

Sticker Paths

If you just want to add some images with a specific category without moving/copying them, you can do that too. The Paths configuration lets you add any number of folders, anywhere on your PC, as stickers for any category you like. For example, with the following configuration:

Stickers:
  Paths:
    Discreet: ["C:/pictures/discreet-stickers"]
    Chastity: ["C:/pictures/chastity-stickers", "D:/downloads/stickers"]

{
    "Stickers": {
        "Paths": {
          "Discreet": ["C:/pictures/discreet-stickers"],
          "Chastity": ["C:/pictures/chastity-stickers", "D:/downloads/stickers"]
        }
    }
}

Any images in any of the folders provided for a given category (Discreet and Chastity in the example above) will be merged together and used any time a client requests sticker censoring with those categories enabled.

You also don't need to worry about images being the exact right dimension! Beta Censoring will check the available images and find one that's aspect ratio is close enough to the censoring it's being used for. While we recommend sticking mostly to square-ish images, you don't need to worry too much about the exact dimensions.

Sticker Packs

You will sometimes find pre-prepared sticker packs. These are special archive files with the .betapkg file extension. You can drop these right into the server folder (i.e. where BetaCensor.Server.exe is) and the server will load any stickers contained in the pack when it starts.

To get you started, here's a couple of basic sticker packs:

Pack	Author	Description	Download
Emoji	silveredgold	A simple pack of emojis. Mostly square, with some mosaics for wider matches.	Download Here
Professional	finalfrog	A more "professional" or "corporate" set of stickers with a mix of square and wide stickers.	Download Here
Labels	silveredgold	A simple set of censoring-themed stylized text labels in mostly wider shapes, particularly useful if your other stickers are mostly square.	Download Here

Startup Mode

By default, Beta Censoring will run in its "Normal" mode which tries to speed up sticker loading by pre-loading your available stickers into memory. This meaks it much faster to apply sticker censoring at the cost of increased memory usage and startup speed. In our ever-present goal for customization, this is configurable. You can use your configuration file to set the sticker startup into either "Hybrid" or "Fast" mode.

Stickers:
  StartupMode: Hybrid

{
    "Stickers": {
        "StartupMode": "Hybrid"
    }
}

Your options are summarized below:

• Normal (default): This mode takes longer to start up and increases memory consumption, but makes censoring (when using stickers) much faster. Adding/removing stickers requires restarting the server.
• Hybrid: This mode still pre-loads stickers, but does so a little bit faster. This mode is a little faster to start up at the cost of slightly slower censoring (when using stickers). Adding/removing stickers requires restarting the server.
• Fast: This mode does not pre-load stickers at all. This means it is the fastest to start up and uses significantly less memory, but is noticeably slower to censor (when using stickers). Adding/removing stickers does not require restarting the server.

I recommend leaving the server in Normal mode unless you are really struggling for memory and are willing to accept a penalty in censoring time when using stickers.

Captions Configuration

Beta Censoring includes a default set of captions so you only need to do this if you want to change them.

Simple Method

The simplest method if you just want to change which captions are shown by default, you can add a captions.txt file in the same folder as the server, with one caption per line. These will replace the default captions.

Advanced Method

If you want a bit more control, or have a more complex set-up in mind, you can configure captions from the configuration file. Captions work similarly to stickers to configure, but with a Captions configuration section that looks a little different:

Captions:
  Captions:
    - "unworthy"
    - "not for you"
  FilePaths:
    - "D:/captions.txt"

{
    "Captions": {
        "Captions": ["unworthy", "not for you"],
        "FilePaths": ["D:/captions.txt"]
    }
}

In short, you can include the captions directly in the configuration file (as a list in the Captions property) or include any number of text files where each line of the text file is one caption (as a list of paths in the FilePaths property).

While the API supports it, the current default caption provider ignores requested categories.

Censoring Configuration

You can also fine-tune the censoring settings used by all clients using the server configuration file. You can separately adjust the minimum confidence (between 0 and 1) for Beta Censoring to censor a result from the AI, as well as tune the scores required for individual matches ('classes' as the AI calls them). For example, the default scores range from 0.4 to 0.55 (the default) up to 0.6 for some of the trickier covered classes. If you want to make the censoring more trigger-happy overall, you can adjust the MinimumScore option, like below:

MatchOptions:
  MinimumScore: 0.4

{
    "MatchOptions": {
        "MinimumScore": 0.4
    }
}

or if you just want to make it much more likely to censor specifically breasts (either covered or exposed) you could use the following:

MatchOptions:
  ClassScores:
    EXPOSED_BREAST_F: 0.35
    COVERED_BREAST_F: 0.45

{
    "MatchOptions": {
        "ClassScores": {
          "EXPOSED_BREAST_F": 0.35,
          "COVERED_BREAST_F": 0.45
        }
    }
}

You can provide class-specific scores for any of the classes supported by the NudeNet model.

Server Configuration

If you're looking to tweak how the server itself works, the most useful configuration block is the Server block. When not specified, the default is equivalent to the following:

Server:
  WorkerCount: 2
  EnableSignalR: true
  EnableRest: true

{
    "Server": {
        "WorkerCount": 2,
        "EnableSignalR": true,
        "EnableRest": true
    }
}

Worker Configuration

It may be tempting to dramatically increase the WorkerCount setting to get more workers censoring images at once, but this is not usually a good idea.

Increasing the worker count will dramatically increase the load on your PC while censoring. Additionally, adding more workers actually has the potential to slow down censoring, especially if you add more workers than your PC can reasonably run at once. A reasonable rule of thumb is to use half the number of cores your CPU has. If you're okay with things really slowing down during censoring, you can try going as high as the number of cores, but it's strongly recommended to not go above this number.

Optimization Mode

Beta Censoring includes a bunch of built-in optimization behaviour to try and keep performance acceptable. The defaults should be a reasonable balance of speed to accuracy, but you can also tell Beta Censoring to either be more relaxed (slower, but more accurate) or more aggressive (faster, but less consistent).

In your configuration you can set the optimization mode to one of the following:

• None: Skips most optimizations. Slower, but more consistent and accurate
• Normal (default): Adds some pre-processing to slightly improve AI performance without sacrificing too much accuracy
• Aggressive: Applies heavier pre-processing and inference to significantly increase the AI performance but with a much higher risk of missed or inaccurate matches.

In your configuration, you can set the optimization mode to one of the above modes if you want to override the default.

Server:
  OptimizationMode: Aggressive 

{
    "Server": {
        "OptimizationMode": "Aggressive"
    }
}

Scaling Configuration

There is one additional configuration item you might be interested in: image rescaling. When Beta Censoring actually runs an image through the AI, it will (if necessary) scale the image down a little to speed things up. By default, though, it will only scale anything larger than 1080p. This default should result in reasonably accurate matching from the AI with reasonable performance.

If you're okay with less accurate matching and want faster processing, you can tell Beta Censoring to scale images down before processing them. That will result in faster but less accurate results from the AI. To do so, change the ImageDimensions option in the server config:

Server:
  ImageDimensions: "1280x720" # note the quotes are required!

{
    "Server": {
        "ImageDimensions": "1280x720"
    }
}

This would result in all images being scaled down to 720p before being handed to the AI. You can specify either a set of <WIDTH>x<HEIGHT> dimensions (as above) or just a single number for both.

Note that censoring (and therefore the final censored image) will always be on the full-size image.

Caching Configuration

Beta Censoring from v0.0.10 onwards includes support for caching results from the AI and censoring process. This means that if you request the same image to be censored, the backend can just grab the results from the cache and return them straight away rather than running the image through the AI and censoring processes again.

Caching is split into two parts: the AI cache (aka the results cache) and the censoring cache. The AI cache just caches the matches that get returned from the AI model (i.e. what body parts are where), while the censoring cache will cache the entire censored image. You can individually toggle these caches so that you can have the AI results cached but still get images newly censored each time (great for stickers and captions users) or enable the censoring cache as well for the fastest possible performance.

Note that enabling either cache will increase the server's memory usage, but this is especially noticeable for the censoring cache which will hold every image you have censored in the last 15 minutes in memory, which can use a lot of memory if you browse a lot of images.

To enable/disable caching, use the separate Caching section of the configuration file:

Caching:
  EnableMatchCaching: true
  EnableCensoringCaching: false

{
    "Caching": {
        "EnableMatchCaching": true,
        "EnableCensoringCaching": false
    }
}

Other Configuration

There is also some extra configuration options to tweak the censoring behaviour available in the CensorOptions section that let you globally control some parts of the matching/censoring process. Of particular note:

• RelativeCensorScale: This will globally scale the size of matches from the AI up or down. A value of 1 means that censoring will only apply to the area indicated by the AI, while (for example) 1.2 will include an area 20% larger in each direction.
• PaddingScale: This can be used to increase/decrease the padding area around matches used by the censoring process. We don't recommend you change this.
• ForcePixelBackground: By default, stickers will be applied on top of a blur effect. Setting this to true will instead pixelate the area under the sticker. This effect is necessary as the server has no way of knowing whether or not the sticker is actually covering the area that is to be censored, so we blur or pixelate the area then apply the sticker on top.
• ClassStrength: This is currently only used by the blur and pixelation censor types, but can be used to increase the strength of those effects for specific classes (i.e body parts). This is multiplicative with the censor level requested by the client, so be careful with it.

# all values are optional
CensorOptions:
  RelativeCensorScale: 1.1
  ForcePixelBackground: true
  ClassStrength:
    FACE_F: 0.9

{
    "CensorOptions": {
        "RelativeCensorScale": 1.1,
        "ForcePixelBackground": true,
        "ClassStrength": {
          "EXPOSED_BREAST_F": 1.2
        }
    }
}