hjLabs AI Playground
Starting the on-device runtime…
OWL-ViT locates objects from a text prompt rather than a fixed label set, so you can ask for "a red safety helmet", "cracked tile" or "barcode" and get boxes back without collecting a single training image. That makes it the fastest way to test whether a detection idea is viable at all before anyone annotates a dataset. Accuracy trails a fine-tuned detector, but the iteration loop is seconds instead of weeks — and it all runs on your GPU.
zero shot object detectionowl-vit demotext prompt object detectionopen vocabulary detectionfind objects by descriptionno training detectiongrounding detection online
Short noun phrases work best — "a person wearing a hard hat" beats a full sentence. Give several candidate phrases at once and compare their scores.
Whenever the thing you are looking for is not one of the 80 COCO classes, or when you are still exploring what to look for. Once the class list is fixed and you need accuracy, fine-tune a real detector.
No. The model weights are downloaded from Hugging Face to your browser once, and every inference after that runs on your own GPU through WebGPU. Your text, images and audio are never sent anywhere — hjLabs.in has no inference server and no way to see your input.
Small models do. Recent mobile Chrome and Safari expose WebGPU, but memory is the binding limit — stay under roughly 200 MB of weights on Android, and expect iOS to reclaim the tab sooner. The catalog shows every download size up front so you can pick something that fits.
That depends on the individual model, not on this site. Each model page links to its Hugging Face repository, where the licence is stated — most of the catalog is Apache-2.0 or MIT, a few are research-only. We add nothing on top: hjLabs.in claims no rights over anything you produce here.