lAIbel is an open-source image labeling platform that works in any browser, locally or in the cloud. Built to speed up dataset preparation for computer vision, it combines a clean GUI with AI-assisted auto-labeling.
Created by Intel® Liftoff startup EnvisionAIry, lAIbel was recently introduced at the After.Databricks() Developer Meetup, part of the Intel AI Summit Networking series. The team shared how they’ve rethought the way developers and researchers approach dataset labeling, turning a slow, manual step into a faster, smarter process.
Based in the open-source community, Envisionary builds tools that simplify complex AI workflows. With lAIbel, they focused on one of the biggest time drains in computer vision: labeling training data. Their answer: a lightweight, browser-based tool that works locally or in the cloud and speeds up annotation with help from AI.
- GitHub: lAIbel on GitHub
- License: MIT
- Author: Dipro Chakraborty, Founder at EnvisionAIry
What Problem Does It Solve?
Building useful computer vision models starts with labeled data. But labeling images, especially at scale, can take hours of repetitive work. Most tools either limit you to one platform, require setup, or can’t
handle AI-assisted workflows.
lAIbel removes those hurdles. It’s lightweight, browser-based, and works across environments. Whether you're labeling locally on your laptop or in the cloud with high-performance hardware, the experience is consistent and fast.
What We Tested Together
To help Envisionary push lAIbel even further, the team joined forces with Intel to validate the tool across a mix of local and cloud-based setups. Together, we explored how well it performs with AI-assisted features, and how it integrates with downstream training workflows.
This collaboration ran through Q2 2025 and included tests on:
- Intel® Gaudi 2 AI Accelerator for cloud-based workloads
- Intel® AI PC for local, private workflows
- Compatibility with YOLO-E and SAM models
- Prompt-based labeling using text input
Tech in Action
Once up and running, lAIbel connects with powerful models to simplify the labeling process. It doesn’t just let you draw bounding boxes, it uses AI to suggest them.
For example, type “glasses” into the prompt box, and lAIbel can identify and tag that object automatically using YOLO-E.
Here’s a snapshot from one of the tests:
- Detection target: glasses
- Inference time (YOLO-E on Gaudi): 0.3993s
- Confidence: 0.950
- Average time (YOLO-E): 165–420 ms
- Average time (SAM): ~50 ms
These numbers show that lAIbel isn’t just easy to use, it’s built to work efficiently, even on large-scale datasets.
Why It Matters
Labeling is often the most tedious part of building AI products. EnvisionAIry designed lAIbel to take that load off your team. By combining a clean UI with AI assistance, it gives researchers, developers, and builders a faster way to prepare training data without being tied to a specific platform or toolset.
Whether you’re in a secure lab with strict data policies or working on cloud-scale AI, lAIbel meets you where you are.
What’s Next
Looking ahead, the EnvisionAIry team plans to expand lAIbel with features like data augmentation and active learning, letting users train models in the loop and refine datasets as they go. This keeps the feedback loop tight and improves model accuracy with less effort.
Try It Yourself
You can start using lAIbel right away. It’s open-source, MIT-licensed, and ready to run.
https://github.com/bortpro/laibel
Resources
I'm a proud team member of the Intel® Liftoff for Startups, an innovative, free virtual program dedicated to accelerating the growth of early-stage AI startups.