If you want a high-level deep learning framework that's easy to use, Keras is a great option. Its API is elegant and easy to maintain, reducing cognitive overhead so you can concentrate on the logic of your project. Keras can run on several backend frameworks, including TensorFlow, JAX and PyTorch, which means it's very flexible and can be used for a broad range of needs. The documentation and tutorials are well written, and there are more than 150 examples to get you started.
Another top contender is PyTorch, which offers a flexible and approachable front-end for rapid prototyping and efficient production. PyTorch supports distributed training, performance tuning and a variety of tools and libraries. Its ability to switch between eager and graph modes with TorchScript and to import and export ONNX format models means it's good for everything from quick prototyping to large-scale production.
TensorFlow is another top option, offering a flexible environment for building and running machine learning models. TensorFlow offers several levels of abstraction, including the high-level Keras API, eager execution and distributed training. It can be used for a broad range of tasks and has tools like TensorFlow Lite for on-device training and TensorFlow.js for web deployment, so it's good for beginners and experts.