We’re excited to announce a major set of updates designed to make your training workflow more powerful, flexible, and easier to manage!

🔹 1. Library now supports Labeling

• You can now create folders in the Library and label your own images / videos directly inside them.

• Folders can be imported into online training with one click, or downloaded for local training.

• Easily organize, manage, and back up all your training assets in one place!

You can now edit labeling directly from the left panel in a more intuitive way.
We’ve also upgraded the labeling component in Model Training, bringing the same convenience to the training page.

🔹 2. New Labeling Models Added

We’ve introduced 3 new labeling models:

• gemini-3-flash

• gemini-2.5-flash

• qwen-3-vl

✨ These models support style preference prompts, allowing you to precisely control labeling results.
👉 We've prepared preset templates for you. You can also freely configure your exclusive Labeling Prompt based on your training set.

📌 Note: Labeling with these three models will consume small amount of credits
You can always switch back to the previous free models if needed.

🔹 3. Upload Local Labeled Datasets Directly to Library

• You can now upload your locally labeled training dataset directly to the Library.

• Simply make sure your images and corresponding labeling .txt files share the same filename.

• Once uploaded, they will be automatically paired and stored in the Library for easy management.

Here is a short video illustration: https://tensor.art/posts/956859563527662579

🔥 This update is a huge boost for model trainers — better organization, smarter labeling, and smoother training workflows.
Try it out now and let us know what you think!