T-Rex Label Review (2026): Pricing, Features & Honest Verdict

TLDR

T-Rex Label is a browser-based AI annotation tool for computer vision. Draw one box, and its T-Rex2 model labels every similar object across your images. Best for: ML engineers and small CV teams. Price: Free interactive labeling; AI pre-annotation is credit-based. Rating: 7.9/10.

What is T-Rex Label?

T-Rex Label is an online tool for labeling images to train computer vision models. You upload a dataset, draw a single bounding box around one example of an object, and the tool finds and labels every other instance of that object across your images. It handles bounding boxes and segmentation masks, runs entirely in the browser, and needs no install, no GPU on your machine, and no model fine-tuning.

The detection engine is T-Rex2, a model published by IDEA Research at ECCV 2024 for generic object detection via text and visual prompts. The hosted product is operated by Vision Future Technology in Shenzhen, with the first Product Hunt launch in October 2024 and a steady cadence of feature drops since: cross-image annotation in May 2025, AI auto-labeling in June 2025, and AI mask annotation in October 2025. The pitch matters because labeling is still the slowest, most expensive step in building a vision model, and zero-shot prompting cuts that work dramatically.

Most annotation tools fall into one of two camps. Either they are heavy MLOps platforms that bundle labeling with hosting, training, and deployment, or they are bare manual labelers where every box is a click. T-Rex Label sits in a useful middle: it is a focused labeling tool, but the AI does the repetitive detection for you. That focus is the point. It does not try to be your whole pipeline, it just tries to get your dataset labeled faster than anything else, then hands the result off to wherever you actually train. We tested it against the kind of dense, repetitive scenes that wreck a manual workflow, and the visual-prompt approach held up.

What Are T-Rex Label’s Key Features?

Visual prompt detection

This is the headline. You draw one box around a target object, and T-Rex Label predicts boxes for all similar objects in the image in a single step. In dense scenes, counting cattle in a field or parts on a circuit board, this turns dozens of manual clicks into one. The model is open-set, so it recognizes objects that were never in its training data, which is the difference between a tool that only knows COCO classes and one that handles your specific factory part.

Cross-image inference

Once you prompt an object in one image, T-Rex Label propagates that prompt across the rest of the batch and auto-labels the same object everywhere. For a folder of a few hundred similar shots, you set up the prompt once and let it run, instead of repeating the same labels image by image. This is where most of the claimed time savings come from.

AI mask annotation

Added in October 2025, mask labeling produces pixel-level segmentation, not just rectangles. It supports both traditional manual mask drawing and an automatic mask mode, and the automatic mode is free for the community. Segmentation is normally the most tedious annotation type, so having an auto path for it is a real differentiator against tools that charge for every mask.

Open-set and zero-shot detection

The reason the visual prompts work on your data and not just on stock categories is the open-set model underneath. Traditional detectors only recognize the fixed list of classes they were trained on, so a custom factory part or a rare crop disease is invisible to them. T-Rex Label’s model recognizes objects outside its training data without retraining, which is what lets a single drawn box generalize. In practice this means you spend zero time fine-tuning a base model before you can start labeling, the gap that usually eats the first day of any annotation project.

Export and platform integrations

Labels are only useful if they leave the tool cleanly. T-Rex Label exports in formats compatible with Roboflow, COCO, YOLO, TensorFlow, Hugging Face, ModelScope, Kaggle, and Label Studio. That means you can do the fast labeling here and still train in whatever pipeline you already use, rather than getting locked into one vendor’s training stack. For teams already invested in a particular framework, this matters more than it sounds: the cost of switching annotation tools is usually the export format, and T-Rex Label removes that friction by speaking everyone’s language.

How Much Does T-Rex Label Cost?

The most important number is zero. Interactive and manual labeling, including cross-image auto-labeling and automatic mask annotation, is free for the community with no image cap we could find. For a solo researcher or a small team labeling images by hand or by visual prompt, you can run the whole workflow without paying.

The paid layer is AI pre-annotation, the fully automated batch labeling that runs models over your dataset without you prompting each object. T-Rex Label charges this on credits, billed by the number of images annotated and the model type used. The honest catch: there is no published plan table or self-serve checkout. You start free and contact the team when you need paid pre-annotation volume, which makes upfront budgeting hard.

For context on value, Roboflow starts at $49/mo and its Core plan runs $79/mo on annual billing with limited monthly credits. Labelbox bills consumption at roughly $0.10 per labeling unit, which adds up fast at scale. CVAT is free but you self-host and maintain it. T-Rex Label’s free interactive tier undercuts all of the paid options for straightforward image labeling, and its credit model only kicks in for hands-off automation.

Who is T-Rex Label Best For?

Use T-Rex Label if you are an ML engineer, data scientist, or small computer vision team that needs to label image datasets quickly without installing or training anything. It shines for object detection and segmentation in agriculture, retail, electronics, medical imaging, and logistics, where you have many near-identical objects per image and want zero-shot detection to do the grunt work.

Skip T-Rex Label if you need to annotate video, 3D point clouds, audio, or text, since the tool is image-only. Skip it too if your organization requires a published, predictable price for automated labeling before you commit, because the credit pricing for AI pre-annotation is opaque. Teams that need a full MLOps platform with hosted training and deployment will outgrow it and want Roboflow instead.

Best T-Rex Label Alternatives

Roboflow

Roboflow is the category leader and does far more than label: it hosts datasets, trains models, and deploys them. Pricing starts at $49/mo, with the Core plan around $79/mo on annual billing and Growth at $299/mo. If you want one platform for the whole vision lifecycle, Roboflow wins. If you only need fast, free labeling and will train elsewhere, T-Rex Label is leaner and cheaper.

CVAT

CVAT is the open-source workhorse of image and video annotation. The self-hosted version is free, which is its biggest advantage, but you maintain the infrastructure and its AI-assist features are weaker out of the box. T-Rex Label is easier to start with and has stronger zero-shot auto-labeling, while CVAT gives you full control and video support that T-Rex Label lacks.

Labelbox

Labelbox targets larger teams with a consumption model around $0.10 per labeling unit plus managed labeling services. It is built for governance, review workflows, and scale across many data types. It is overkill and expensive for a small team labeling images, where T-Rex Label’s free interactive tier does the same core job for nothing.

Final Verdict: Is T-Rex Label Worth It?

For image annotation, T-Rex Label earns its 7.9. The visual-prompt workflow built on T-Rex2 is genuinely fast, the export support means you are not locked in, and the fact that interactive labeling, cross-image auto-labeling, and automatic mask annotation are all free puts it ahead of every paid alternative for straightforward object detection work. We found very little to dislike about the actual labeling experience.

The reservations are about commercial clarity, not capability. AI pre-annotation runs on credits that are not publicly priced, there is no plan table, and the tool is image-only with no video, 3D, or text. The opaque credit pricing is the thing that would give a procurement team pause, and it is the main reason this scores 7.9 rather than higher. If those gaps were closed, a published pre-annotation tier and any video support, this would be a top-tier pick outright. As it stands, the free interactive labeling is so strong that the missing pieces only matter once you scale into heavy automation.

If you label images and want speed without spend, start here today and only reach for Roboflow or Labelbox when you need a full platform or strict procurement-friendly pricing. For most small and mid-size CV teams, T-Rex Label is an easy yes.