Cadryl: Image Annotation

Your images, your vision. Cadryl brings import, annotation and export into one Android workspace. Prepare image datasets, review each image and stay in control of your annotations.

Annotate your way
Set up your classes and tasks. Draw boxes, place points, create masks and write captions. Work manually or edit suggestions from a compatible model. Saved human annotations stay protected during model operations.

From Hugging Face to Hugging Face
Import a local folder or a Hugging Face source. Organize images into batches, approve or reject cases, then prepare a local export or explicitly publish to Hugging Face. Copy checks support batch completion.

Formats for your workflow
Export JSONL, COCO, YOLO, WebDataset or vision-language data according to your task and annotations. Keep a complete workspace backup to recover projects, annotations and progress. A dataset export is not a complete project backup.

Models on your device
Import compatible bundles and run local inference. With a model that supports training, train a separate version using approved annotations and choose whether to activate it. The original remains available. Capabilities and performance depend on the model and device.

For your computer vision projects
Prepare datasets, review a collection or explore image annotation on your phone. An interactive tutorial helps you get started. The app interface is available in French and English.

Your personal API
Connect a provider and vision model compatible with the OpenAI protocol using your own key. Confirm each image transfer and review suggestions before accepting them; your human annotations remain protected. The key is encrypted on the device and excluded from backups and exports. Cadryl does not host models or include API credits. Your provider bills usage separately.

No ads or subscription
Import, annotation, correction, export, compatible models, batch annotation and compatible training are available without an in-app purchase. The studio also has a tablet-optimized interface.

Before you start
Android 9 or later and an ARM64 device are required. No model weights or datasets are bundled. Use images and models for which you have the necessary rights. Projects and local processing stay on the device. Hugging Face exchanges, downloads and API calls use the external services you choose. Cadryl has no backend of its own. See the privacy policy for details.

Recent changes:
Fixed Hugging Face setup: access images when the Viewer fails, choose the source folder and use private repositories with an authorized token.