A fork of coktailpeanut/breadboard designed to improve image captioning workflow. This is an untested early alpha. Use at your own risk.
Add your image dataset to the Connected Folders section in settings and Re-Index It will add any existing caption files to the database, allowing searchability. Captions are now searchable, and images can be filtered by whether they have a caption or not:
The caption wizard can be accessed in the toolbar:
It will apply captioning to any selected images, or all images visible within the current filter if none are selected.
- Max Caption Length Maximum tags to use
- Existing Caption Action How to handle existing caption if present
- GIT Pass Whether to perform a pass using Microsoft's GIT model
- GIT Fail Phrases Don't use GIT caption if it contains one of these phrases
- BLIP Pass Caption using BLIP model if selected and/or GIT fail phrase detected
- BLIP Beams Number of beams for BLIP pass
- CLIP Min/Max Length Number of tags to use from CLIP
- CLIP V2 Model Currently broken, can ignore
- APPEND (Artist/flavor/method/movement/medium) Tags from CLIP If any of these are selected, will perform a CLIP tag pass
- Tags to ignore Comma seperated, will not add matching tags to caption
- Replace class with subject in caption Will perform a find/replace on subject class and replace with subject name after tags are generated
- Use folder as tag Will tag the image with the parent folder names. Useful if you have images grouped by artist/style/subject
- Folder levels to tag Number of nested parent folders to add as tags
- Write caption to file Will only write to txt file if checked. Otherwise, it will just display the new caption in the status window. Useful for doing preview runs without modifying existing caption files
- Use filename as existing caption Will parse the filename to use as the existing caption
Clone this repo
run pip install -r requirements.txt
run npm install
run npm run