batch tagging workflow

How to Use the Workflow

The nodes in this workflow are clearly organized. You just need to configure the following key parameters in order to start the batch processing:

Step 1: Input Training Set Folder

• Specify the folder path containing all the images you want to tag.

Step 2: Select Model Type

• Choose the appropriate recognition model based on your image content (e.g., General, Anime) to get more accurate tags.

Step 3: Select Tag Type

• (Core Feature) This determines the style and format of your generated tags. Choose based on your training needs:

  • Descriptive / Straightforward: Generates descriptive sentences.

  • Danbooru / Booru-like tag list: Generates comma-separated keywords in the Danbooru style, which is excellent for anime-style training.

  • MidJourney / Stable Diffusion Prompt: Generates tags styled like MJ or SD prompts.

  • Many other types like Art Critic and Social Media Post are available for you to explore.

Step 4: (Optional) Advanced Content Control

• (Fine-tuning) This section provides a series of toggles to precisely control what content to include or exclude, allowing for high-level customization. For example:

  • include_lighting: Adds a description of the lighting to the tag.

  • include_camera_angle: Adds a description of the camera angle.

  • exclude_people_info: Excludes tags related to people's information.

  • include_character_age: Adds the character's age.

  • You can freely combine these options to generate tags that best suit your needs.

Step 5: Select Tagging Length

• Set the desired length for the generated tags. You can choose between short, core tags or more detailed descriptions depending on your training requirements.

Step 6: Longest Edge

• This is an image preprocessing parameter. Keeping the default value is usually sufficient; no changes are typically needed.

Step 7: Input Style Trigger Words

• Enter any general words or trigger words you want included at the beginning of every tag. For example, when training a LoRA for a specific character, you can input the character's name here, like character_name.

Step 8: Translate to English

• (Key Feature) The workflow has a built-in translation node. By enabling it, you can automatically translate the generated tags into English. This is a necessary step for most model training and can now be done all at once!

Step 9: Input Output Folder Path

• Select a folder to save all the final tag files. The workflow will automatically generate a .txt file for each image with the same name.

Important Notes

1. Assistant Tool, Not a Full Replacement: Please remember that AI auto-tagging is not 100% accurate. This workflow is intended to provide a good starting point. The generated tags must be manually reviewed and corrected to ensure the quality of your final training set!

2. Recommended Process: Batch Generate -> Manually Refine -> Finalize Tagging. Do not skip the "Manually Refine" step!

3. Check the Output: After the run is complete, please check your designated output folder to confirm that the .txt English tag files have been successfully generated for your images.

We hope this enhanced workflow helps you significantly improve your dataset creation efficiency!

Feel free to download, try it out, and share your feedback!