TagScribeR

Proof of concept for a UI to streamline AI image captioning

Introducing TagScribeR, the ultimate toolset designed to revolutionize the way you prepare datasets for AI-driven image generation. Our cutting-edge platform not only streamlines the tagging and captioning process but also brings a new level of precision and personalization to your Stable Diffusion models. Say goodbye to the tedious manual tagging and hello to seamless, AI-powered auto-captioning and intuitive, user-defined tagging. With TagScribeR, crafting your perfect dataset is only a few clicks away, saving you time and unleashing your creative potential like never before.

TagScribeR: Redefining AI-Driven Image Dataset Preparation

In the burgeoning field of AI and machine learning, the quality and detail of training datasets are paramount. TagScribeR is a pioneering platform designed to address the challenges of dataset preparation for Stable Diffusion models. Our innovative tool leverages the power of AI, specifically BLIP-2 utilizing CLIP Interrogator, BIG THANKS Mr.Pebble in Discord and an EVEN BIGGER THANKS to the CLIP Interrogator team for making the implementation of auto captioning possible, to auto-caption images, while providing a user-friendly interface for custom tagging.

Key Features:

Auto-Captioning with BLIP-2: Harness the power of AI to generate accurate, context-rich captions for your images, laying a solid foundation for your dataset. Custom Tagging Made Easy: With our intuitive interface, add personal touches to your dataset by appending custom tags, ensuring your model's output aligns perfectly with your vision. Seamless Dataset Organization: Sort and manage your images effortlessly, turning the chaotic sea of files into a well-organized library tailored to your needs. Community Collaboration: Share and utilize tag sets within the TagScribeR community, fostering collaboration, and standardizing tagging practices across projects. Why TagScribeR?

Save Time, Amplify Creativity: Reduce the hours spent on dataset preparation and invest more time in what truly matters - your creative and innovative outputs. Precision at Your Fingertips: With AI-powered captioning and personalized tagging, your datasets will reach new heights of relevance and precision. Community and Support: Join a growing community of professionals and enthusiasts. Share, learn, and grow together with shared resources and support. TagScribeR is not just a tool; it's your next step towards mastering AI-driven image generation. We're seeking supporters, backers, and collaborators who share our vision to transform the landscape of AI model training. Be part of this revolutionary journey with TagScribeR – where your creativity meets AI efficiency.

Join Our Mission: Collaborate, Back, Code! 🚀 We're Building Something Big – Be a Part of TagScribeR

TagScribeR is more than just a tool – it's a vision for a future where creators and technologists can seamlessly merge their talents to train AI models with unmatched precision and personalization. We believe in a world where the power of Stable Diffusion models is unlocked by the collective creativity of a community, and where every dataset tells a vivid story.

We Need You!

👨‍💻 Collaborators: If you are passionate about AI, image processing, or UI/UX design, we need your insight and expertise to bring TagScribeR to life. Whether you're a seasoned pro or looking to contribute to a meaningful project, there's a place for you here.

💼 Backers: Visionaries and investors, your belief in a project can be the wind beneath its wings. By backing TagScribeR, you're not just funding a tool; you're investing in the future of AI-driven content creation. Help us turn this dream into a reality.

🧑‍💻 Coders: We're on the lookout for talented developers who are ready to dive into the exciting world of machine learning, front-end, and back-end development. Your code could be the cornerstone of a new era in Stable Diffusion model training.

Why Get Involved?

🌟 Innovate: Be at the forefront of technological advancement. 🤝 Collaborate: Work with a team of motivated individuals driven by passion and the urge to create. 📈 Grow: Gain exposure, experience, and the satisfaction of contributing to a groundbreaking project. 💡 Create Impact: Your contribution could revolutionize the way AI models are trained and how creative content is generated. How Will Your Role Shape the Future?

Every collaborator brings a unique perspective that could pivot the project towards success. Every backer adds a brick to the foundation of our development. Every coder stitches a line of code into the fabric of TagScribeR’s legacy.

Ready to Make a Mark?

Join us on this thrilling journey. Check out the rest of our README to understand our vision and direction. If you're feeling inspired and see a potential spark, get in touch! Let's explore how your skills and ideas can synchronize with TagScribeR.

Reach Out Now!

Whether you have a question, suggestion, or ready to jump in — we're here to listen. Connect with us at [email protected], or jump straight into our issues and pull requests. Let's create, innovate, and elevate together.

TagScribeR is waiting for you. Let's tag the future, together.

TagScriber v0.03.0-alpha

Current Build is meant for Windows OS and Python as that is what I am working with.

Current features:

• Load Directories and display/edit associated text files in Gallery tab/widget

• Load Directories with blank text fields meant for populating with Auto-Captions once BLIP-2 is integrated

• Image selection (Multi-Select images functionality)

• Tag panel (collapsible tag panel with text field to add/create new tag buttons and also delete tag buttons if needed)

• Tag Buttons (functions to create tag buttons and functions to append current text/captions to selected images when clicking a tag button)

• Persistent Tag buttons (user created tags persist from session to session unless deleted)

• Sort/Filter function (Sort/Filter by tags with displayed images & text being updated in the window to display only the images/texts associated with the tag being searched/filtered for)

• Category/Collections Pane/Panel & Functions (Custom categories/collections pane to create new directories/dataset folders to copy selected files into for an easy way to mix n match and create new datasets)

• Tag Sharing (a text file is created in your root folder housing all user custom tags, this text file acts as your tag button database but also allows easy sharing of user made tags)

• BLIP-2 integration (Integrate Blip-2 and Blip-2 settings to auto generate captions to text fields and save generated captions to new text files named after target images) (CPU Only based currently)

• Image Editing (Added an Image Editing Widget powered by OpenCV/cv2 for image editing, current image editing features are rotation and resizing, plan to add more soon)

• Settings (Added a settings widget to house current and future program settings. Currently there is a guide for hotkeys and the ability to change from Default/Light theme to Dark Mode.)

• Keyboard Hotkeys/Shortcuts - Added keyboard shortcuts to the Gallery Window, Also the Shortcuts guide is now available in settings. Remapping Now Available! Note: Gallery uses all of these; Auto Captioning only uses Save, Undo/Redo, Select/Deselect, Clear Selected Captions; Image Editing only uses Save, and Select/Deselect All.

Ctrl+S: Save all edits

Ctrl+Z: Undo last action

Ctrl+A: Select/Deselect all images (toggle)

Ctrl+F: Focus on search bar

Del: Clear selected captions

Ctrl+L: Load directory

Ctrl+C: Copy selected images to collection

• Added Quick install & Quick launch .bat files (See instructions below)

• Metadata Editor - Metadata Editor is functional, double click to edit fields (still in testing and needs UI adjustments)

• Tag Search, Filter, and Organization - Implemented tag search, filter, and organizing. Tags are now sorted alphanumerically and can be organized into categories.

• How to use:

1. IMPORTANT PREFACE: The way tags.txt is handled has changed. If your previous tags have disappeared, DO NOT PANIC, they are still in your tags.txt in the TagScribeR main directory. The previous method saved the tag in the txt normally, but due to sorting/categorization enhancements, uncategorized tags will need "tag:" prior to the tag. For example: "tag:blonde hair". Simply add "tag:" to the beginning of each tag in the tags.txt file. I recommend using an editor like Notepad++ with the Shift+Alt+Down function to quickly Ctrl+V the "tag:" for multiple lines. Reload the program to see your tags restored and ready for sorting. Note: I tried to take into account the previous tags, and the program should do all this for you automatically, but this is a workaround just in case.

2. Inside Organization Mode:

   • Tick the organization mode box. This will temporarily pause tag button functions.
   • Right-click an empty area in the tag button pane. A context menu will pop up prompting to create a category.
   • Name your category and it will be automatically added to the pane alphanumerically.
   • To sort your tags into a category, right-click the area highlighted next to the tree line (not the button itself). This will prompt a few options: "Move to Category", "Rename Tag", "Delete Tag".
   • For sorting purposes, select "Move to Category". A window with a dropdown will appear for you to choose the category.
   • Select the appropriate category you made and the tag will be sorted into the category. The category is now expandable (tick box non-functional, use the arrow) and your sorted tags are now within your custom category.

3. Outside of Organization Mode:

   • Typing a tag you've saved into the tag search bar will filter through any saved tags that match whatever you type. (If your tag is sorted into a category it will show the category. Select the expansion arrow and it will only show the tags with the words you're searching/filtering for)
   • When a dataset/directory/collection that is tagged/captioned is loaded, right-click on the tag button in your tag pane and it will hide all images with captions that do not match that tag. (I'm working on reorganizing the window to show them uniformly, but for now, the hidden images/captions leave a blank space.)
   • Note: Manual typing and searching of captions/tags is still available at the top for a broader scope of search/filter. The tag filter of images/captions is mainly for jumping to items containing your custom tags of interest.

To do list:

• Image Editing suite (complete Image Editing widget with batch functions to edit/manage/delete/crop/resize/adjust brightness/contrast/hue/color/etc)

• Metadata Editor (Adjust Metadata Editor UI and test all functions thoroughly)

• Separate Image Preview Implement a secondary window or dock-able/resizable window for inspecting individual images.

• Drag and Drop Organization Implement drag and drop organization/grouping for user convenience/comfort/organization while creating datasets.

• Duplicate Detection Implement duplicate detection, and additionally the ability to delete selected images and/or duplicates from the loaded collection/directory.

• Multi-language Support Implement support for most major languages.

• Cloud Integration Implement integration to allow working in/from a cloud storage space (Not super confident in my ability to pull this off, but would be nice if I can).

• Custom Themes/Backgrounds/Widgets/Panes Implement the ability for users to customize themes and backgrounds and sizes and placement of panes/windows, and allow Widget Tab drag reorganization.

Installation and Running

For Windows Users:

1. Ensure you have Python 3.7 or later installed on your system.

2. Git clone to your desired location

3. Double-click the install.bat file to set up the virtual environment and install dependencies.

4. Once installation is complete, double-click the launch.bat file to run TagScribeR.

Manual Installation:

If you prefer to install manually or if you're having issues with the batch files:

1. Create a virtual environment: python -m venv venv
2. Activate the virtual environment:
   • Windows: venv\Scripts\activate
3. Install requirements: pip install -r requirements.txt
4. Run the application: python main.py

License

This project is licensed under the AGPL-3.0 license - see the AGPL-3.0 License file for details.