Builder Tool
The Builder Tool lets you create and expand datasets directly within Bitcrush Studio.
You can import existing Hugging Face datasets, collect media from supported URLs, or create a new dataset from selected images in an existing one.
Bitcrush Studio uses yt-dlp and gallery-dl through their command-line interfaces to download media from supported websites.
Studio installs the required tools automatically. However, you remain responsible for:
- Configuring
gallery-dlfor the services you use. - Providing any required API keys, cookies, account credentials, or authentication details.
- Following the usage limits and terms of each source.
- Keeping your configuration and credentials secure.
See the official gallery-dl documentation for more information about its configuration files and authentication options.
Source permissions
You are responsible for ensuring that you have permission to download and use any imported material for your intended purpose. Bitcrush Studio does not grant licences for third-party content and is not responsible for how web-based sources are used.
Studio supports importing datasets from Hugging Face through the Hugging Face Datasets library.
When you select a dataset, Studio attempts to detect the relevant columns automatically, including:
- Image columns.
- Caption or text columns.
- Tags and other supported metadata.
You can override the detected columns when the automatic selection does not match the structure of the dataset.
This allows fully captioned datasets to be imported directly into Studio while preserving their existing captions and supported metadata.
The URL importer can collect images and videos from supported websites.
Studio uses gallery-dl for most image-based sources and custom integrations where required. For video downloads, Studio first attempts to use yt-dlp. If the website is not supported by yt-dlp but is supported by gallery-dl, Studio falls back to gallery-dl.
For each source, you can control how much content is downloaded and retained.
Available options include:
- Download count: Limits the number of files collected from the source.
- Keep count: Limits how many of the downloaded files are kept in the final dataset.
- Minimum megapixels: Excludes images below a chosen image resolution.
- Automatic tags: Applies selected tags to incoming files.
- Source metadata: Extracts and stores supported metadata provided by the source.
For example, you can download 200 candidate images from a source but keep only 100 in the resulting dataset.
This gives Studio a larger pool to work from without requiring every downloaded file to become part of the final dataset.
Carve creates a new dataset by selecting images from an existing source dataset according to your chosen filters.
This is useful when you want to extract a more focused dataset from a larger collection, such as:
- Images matching particular tags.
- Images within a selected resolution range.
- Images with particular scores or properties.
- A smaller subset of a broader dataset.
The source dataset remains the basis for the operation, while the matching images are used to build the new carved dataset.