This section is about other tools/programs that can be useful for training purpose. They are not part of One Trainer solution, some were introduced in the other-tools Discord thread.

Feel free to suggest any others in the Wiki discussion on Discord.

Dataset Management

Captioning UI's

TagGUI

Tag manager and captioner for image datasets. You can also join their Discord Thread

Features:

• Keyboard-friendly interface for fast tagging
• Tag autocomplete based on your own most-used tags
• Integrated Stable Diffusion token counter
• Batch tag operations for renaming, deleting, and sorting tags
• Advanced image list filtering
• Automatic caption generation with models including CogVLM, LLaVA, moondream, and more
• Option to load auto-captioning models in 4-bit for reduced VRAM usage

Dataset Helpers

Features:

• More specific focus on tag captioning compared to taggui, downside is less models: Supports SW-v3, JoyTag and WD1.4
• Image Cropping, resize, sorting and sequential renaming
• Uses only ONNX models meaning you can run on CPU
• Tag redundancy removal and experimental merging
• Slightly better filtering

JoyCaption Alpha Two

Caption tools using Joy Caption.

• Unofficial GUI mod by D3voz
• Hugging Face Spaces
• Authors Discord Thread
• OT Discord Thread

Features:

• 4bit model support for lower VRAM (D3voz).
• Batch processing

Notes:

• In (D3voz) version prompt options can be edited in gui_updated.py (section EXTRA_OPTIONS_LIST), it updates the UI and the command sent to the model but as Alpha Two is not a general instruction follower, it will not follow prompts outside its training data well. Use this feature with caution.
• Do not support webp format by default (D3voz) but you can enable it in gui_updated.py (load_images class, image_extensions constant).
• I've maded some modifications on it to support webp format, exclude masked images from the image selection and add an option to generate captions as new lines. You can get the files here (discord).

Image Manipulation

Birme.net

Online solution to crop images to a specific resolution.

Image Cropper

Automatically crops images to the closest aspect ratio from a defined list that fits the subject (face, subject).

• Github
• Discord Thread

Testing Models - Prompt helpers

TIPO

Model that will take an input prompt and spit out allegedly improved prompt. Works with Comfy, Forge and A111

Master Prompt

Based on A1111 Unprompted extension. It can generate robust and various prompts for SD1.5, SDXL and SDXL Turbo.

• Discord Link
• Civitai

Features:

• Many generation options such as image type (realistic, digital art), resolution, SFW or not, ethnicity, age, gender, ...
• Can include one or several models (embedding, Lora)

Notes:

• As it's based on an A1111 extension, it's not supported by all solutions. An comfyUI support has been announced but I didn't test it.