BooruDatasetTagManager v2.6.3

• Fixed reopening of the AutoTagger form after cancellation or closing. #222
• Fixed deletion of thinking block in AutoTagger.
• The program update description is now displayed in a separate window. #229

BooruDatasetTagManager v2.6.2

• Added support for video file analysis using OpenAI Compatibility Endpoints

Important! This feature is implemented by obtaining N images from a video (specified in the settings) and transmitting them to the server. This requires quite a few tokens; for example, 50 frames from a 1280x720 video requires ~45,000 tokens. If you use paid platforms, you'll go broke:). If you use LM Studio or similar local programs, you'll need to override the maximum token limit, as the default is 4096 tokens.

BooruDatasetTagManager v2.6.1

• Fixed broken ApiApiServer. #220

BooruDatasetTagManager v2.6.0

• Added support for generating tags using OpenAI Compatibility Endpoints.
• Added progress for loading dataset and autotagging.
• Added display of currently shown images to the dataset. #212
• Added support for loading a dataset with advanced settings. In the advanced settings, you can disable loading preview images, configure the preview image size, and enable reading tags from image metadata (ComfyUI is not supported because the image stores the workflow and it's impossible to determine where the prompt is stored). The code for reading metadata can be taken from this repository. #213
• AiApiServer now supports Qwen3-vl models.
• Fixed a program crash when reading video files. #211 #216
• Fixed missing transparent background when using RMBG (WebP and PNG only) #218
• BDTM now uses NET Core 8.0

BooruDatasetTagManager v2.5.2

• Fixed crash when reading a broken video file (in ScreenLister). #211
• The speed of obtaining images from videos has been increased.
• Fixed dark color scheme of AutoTagger form. #210
• Added huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated model.

If you receive an error when opening the Autotagger settings form, delete the ColorScheme.json file from the program folder and restart the program.

BooruDatasetTagManager v2.5.1

• Fixed background removal tool. #209
• Added support for internlm/CapRL-3B model.

BooruDatasetTagManager v2.5.0

• Added support for video files (mp4, flv, mkv, ts, avi) (Thanks HDANILO).
• Added manual image cropping. (example)
• Added closing forms using the Escape key.
• Added filtering warning in tag replacement form.
• Added to lowercase button in add tag form.
• Added completing text input without using tag autocompletion with Alt+Enter. #205
• Added automatic data substitution in the tag replacement form when two tags are selected. (example)
• Added asynchronous loading of dataset.
• Improved transparent background replacement.
• Fixed caching empty translations.
• interrogator_rpc was partially rewritten, the server part (gRPC) and protobuf serialization were replaced with flask_restful and JSON serialization. Due to interrogator_rpc going beyond the usual tagger, it was renamed to AiApiServer.
• BDTM and AiApiServer now support video autotagging. Models that support video processing are highlighted in a special color. Video processing only supports local paths, i.e. AiApiServer must be located in the same place as the video files, so you won't be able to run AiApiServer on another computer.
• Information about the model repository and video file support has been added to the autotagger settings form.
• Added the ability to specify your own system prompt for models with chat template (Qwen, llama, etc.) (Settings -> AiApiServer -> Custom system prompt)
• Fixed the DeepDanbooru model. (added threshold setting) #207
• Fixed autotagger stopping when there was an error processing an image/video. #199
• Added support for Qwen2.5 based models (Qwen/Qwen2.5-VL-3B-Instruct, Qwen/Qwen2.5-VL-7B-Instruct, huihui-ai/Qwen2.5-VL-7B-Instruct-abliterated, unsloth/Qwen2.5-VL-7B-Instruct-unsloth-bnb-4bit, prithivMLmods/DeepCaption-VLA-7B)
• Added support for NeoChen1024/llama-joycaption-beta-one-hf-llava-FP8-Dynamic model.
• Added support for fancyfeast/llama-joycaption-beta-one-hf-llava model. #198
• Added support for Kwai-Keye/Keye-VL-1_5-8B model. (This model requires flash_attn and triton. For more information, see the readme.)
• Minor fixes and improvements.

Dependencies in AiApiServer have changed, so you need to run pip install -r requirements.txt

BooruDatasetTagManager v2.4.1

• Added support for thwri/CogFlorence-2.2-Large model. #181
• Fixed error that occurred when previewing a previously viewed image. (When caching is enabled) #178

BooruDatasetTagManager v2.4.0

• Added visual tag editor for multiple selection and multiple selection editor. (Preview)
• Added displaying the number of tags in selected images when hovering over a tag.
• Added zooming of preview images in the dataset.
• Added caching of open images (can be disabled in settings).
• Added support for joycaption and moondream2 models.
• Added support for MiaoshouAI/Florence2 models.
• Improved autotagger selection window and added new settings.
• Added tool for cropping images with moondream2 model. (Menu Tools)
• Added tool for background removal using BiRefNet. (Menu Tools and context menu on single image)
• Fixed loading dataset when translation enabled
• Fixed bug that interrupted the autotagger if the model did not contain additional settings. #169
• Fixed List desynchronization detected error when autocomplete is closed with the tab key. #177
• Fixed loading bad (broken) images.
• For convenience, the application was packed into a single file.

Dependencies in interrogator_rpc have changed, so you need to run pip install -r requirements.txt

BooruDatasetTagManager v2.3.1

• Fixed color scheme in autotagger tabs #164
• Fixed threshold value display when interface scaling is greater than 100% #164
• Fixed a bug that caused "List desynchronization detected!" if the selected tag in single selection mode was edited in multi-select mode and then saved without exiting from multi-select mode. #161
• Clarified library dependencies for interrogator_rpc (requirements.txt)