People Tool
The People Tool helps you find, organise, and label the people who appear across your dataset. It discovers names in your captions, tags, and filenames, groups the images each person appears in, lets you record details about them, and writes consistent person tags back onto their images so training data stays tidy and searchable.
Note: The People Tool is a subscriber feature. It is available to supporters with an active Patreon membership; link your account in Settings to unlock it. It is also an early, still-evolving tool — some capabilities, such as face recognition, are on the roadmap and not yet available (see Faces). Everything described below concerns the current, text-based workflow.
The People Tool does not analyse image pixels. Instead, it reads the text already associated with each image and looks for names. Select Scan For People in the toolbar to scan the current dataset across three sources:
- Captions: Names written in your caption text.
- Tags: Names added as tags through Bitcrush Studio.
-
Filenames: Names embedded in image filenames, such as
Anya_Taylor-Joy_001.jpg.
Studio matches names against a bundled dictionary of first names and surnames, and applies several safeguards so ordinary words are not mistaken for people. Descriptive phrases such as "Long Hair", place and award names such as "San Diego" or "Golden Globe", and other common false positives are filtered out. Full names are kept intact, including middle names and hyphenated surnames.
When a scan finishes, each discovered name is handled in one of two ways:
- If the name already matches a person in your People list, the matching images are added to that person as pending matches for you to review.
- If the name is new, it appears as a candidate in the Detected tab.
Note: Filenames are only ever read in the background. The scan reports the names it found and how many images matched, but never exposes the underlying file paths, so filenames that contain private information are not surfaced in the interface.
The Detected tab lists names that were found during a scan but are not yet in your People list. For each candidate you can see the name, which sources it came from (Caption, Tag, or Filename), and how many images it appears in.
For each candidate you can:
- Preview: Inspect the matched images before deciding, which is useful for confirming that an ambiguous name really is a person.
- Accept: Promote the candidate into a full person. Its matched images become pending matches on that person's profile.
- Reject: Discard the candidate.
Clear All removes every candidate from the list at once. A later re-scan may re-detect names, so clearing is a way to tidy the list rather than a permanent block.
The People tab shows every person you have added or promoted, as a grid of cards. Select Add New Person at any time to create a person by hand, without waiting for a scan.
A person can carry a set of optional details, including their birth date, gender, ethnicity, skin tone, hair and eye colour, occupation, and nationality. When NSFW attributes are enabled in your settings, additional physical attributes become available. You only need to fill in the details you want; blank fields are simply left out.
Open a person to see their profile.
A person's profile presents a portrait, a running count of their Accepted, Pending, and Rejected images, an attribute panel, an alias editor, and their image galleries.
The profile toolbar provides several actions:
- Edit: Change the person's name and details.
- Enrich: Look the person up on Wikidata and propose attributes for review (see Enrichment).
- Search Dataset: Re-scan the current dataset for this specific person. Because you have vouched for them by adding them, this search recognises their name even when it is not in the bundled dictionary, and it also matches any aliases you have recorded.
- Sync: Write the person's tags onto their accepted images (see Person Tags).
- Delete: Remove the person. Their image links are removed, but any tags already written to images stay in place.
Aliases let you record alternative names a person may appear under. Aliases are matched during Search Dataset, so a person can be found even when a caption or filename uses a different spelling or stage name.
A person's images are split across two tabs:
- Pending review: Matches Studio has proposed but you have not yet confirmed.
- Approved: Matches you have accepted.
Work through the pending matches by approving the ones that are correct and rejecting the ones that are not. You can approve or reject images one at a time, select several and act on them together, or use Accept All / Reject All to clear the whole queue at once.
A This dataset / All datasets toggle controls whether the galleries show only the images in the current dataset or every image the person appears in across your datasets. Because a person can span several datasets, bulk actions such as moving, tagging, or deleting selected images are routed to the correct dataset automatically.
You can adjust the thumbnail size with the Zoom control, and open any image in its full Detailed View to inspect its captions and tags.
Enrichment looks a person up on Wikidata and gathers structured attributes such as gender, birth date, ethnicity, nationality, and occupation.
Enrichment never changes a person's details on its own. It returns a proposal that you review, so you can choose which suggested fields to apply and which to ignore. Results are cached, so re-enriching the same person does not repeatedly hit the network.
Select Enrich on a single person, or Enrich All from the toolbar to look up everyone in your People list in one pass.
The point of recording people is to get consistent tags onto their images. When you accept an image for a person, assign one directly (see below), or run Sync, Studio writes that person's tags onto the image.
These tags can include:
- The person's name as an identity tag.
- Their recorded attributes, such as hair colour, eye colour, and ethnicity.
- A nationality adjective and their occupation, where known.
- Optional per-image context tags derived from a year found in the filename or caption, such as the person's age at the time of the photo, an age bracket, and the decade the photo was taken.
Tag writing is additive: syncing only adds tags, and deleting a person later does not strip tags that were already written. These tags feed the rest of Studio, so once a person is tagged you can filter and sort their images in Gallery View and pull them into a focused dataset with Carve.
You do not have to run a scan to tag someone. When you open an image's Detailed View in Gallery View, an Add Person field lets you assign the image to any existing person directly. This creates an accepted link and writes that person's tags straight away.
Any people the image is already linked to are shown as chips above the field, and you can select a chip to jump to that person's profile.
The Faces tab is reserved for facial recognition, which is on the roadmap but not yet available. For now, people are discovered from caption text, tags, and filenames using Scan For People, as described above.