ChatTag: Bringing ChatGPT Vision to Image Annotation in OpenFilter

Smarter, Faster, and Built for Real Work

At its core, ChatTag uses ChatGPT Vision to bring human-like perception to computer vision workflows. It doesn’t just detect “a thing in a frame”—it interprets what that thing is, how confident it is about that interpretation, and how it fits into your project’s schema.

This makes it highly adaptable across domains. You can use the same filter to:

• Identify produce in a grocery dataset
• Tag vehicles on a busy street
• Annotate industrial parts for quality control
• Label medical images for diagnostic models

Because the prompt is configurable, you can tune ChatTag for whatever annotation goals your use case demands. It’s a flexible, plug-and-play solution that fits directly into the Plainsight ecosystem.

What Makes ChatTag Different

ChatTag isn’t just another image labeler—it’s a bridge between vision and understanding. Here’s what sets it apart:

• AI-Powered Annotation: Powered by ChatGPT Vision, it can classify and describe objects with human-level reasoning.
• Domain Agnostic: Works on any image—from food to machinery to medical scans.
• Custom Prompts: Define exactly how you want annotations to behave with your own prompt files.
• Dataset Generation: Export ready-to-train datasets in COCO, JSONL, or binary classification formats with a single setting.
• Cost Optimization: Resize or compress images to reduce API usage while maintaining quality.
• Real-Time Visualization: Monitor annotations live through a built-in web interface.

In other words, ChatTag takes the power of multimodal AI and turns it into a practical, production-ready tool for teams that need high-quality labeled data at scale.

Flexible Modes for Any Workflow

ChatTag is designed to fit how you work—not the other way around. You can run it in several modes, depending on your goals.

1. Standard Annotation Mode
   This is the bread and butter. For each frame passed through the filter, ChatTag analyzes the content and attaches structured annotations to the metadata. You’ll see everything from object presence to bounding boxes, complete with confidence scores and token usage stats.
2. Dataset Generation Mode
   If you’re building or expanding a training dataset, enable save_frames. ChatTag will automatically export multiple dataset types—binary classification, COCO detection, and JSONL—so you can jump straight into model training without extra formatting work.
3. No-ops Testing Mode
   Need to test your pipeline without hitting the API? Set no_ops=true, and ChatTag will simulate responses with default annotations. It’s a fast way to validate configuration and performance before running real inference.

Easy to Configure, Easy to Scale

ChatTag’s configuration system is simple but powerful. All parameters can be set in code or through environment variables—perfect for automated deployments and containerized environments.

Example: