Moondream (https://moondream.ai/c/playground and documentation: https://huggingface.co/vikhyatk/moondream2) reads and detects elements of an image (Visual Language Model).

In the “playground” area, four modes are offered:

• Query to question the model,
• Caption for a description of the image in three formats,
• Point to ask the model to point to an element,
• Detect to make the model identify one or more elements.

In English with French accepted in queries. For “Point” and “Detect”, best results with English queries.

Unlimited and open-source.

Images for testing:

• A man wearing face mask cycling by Wat Phra Kaew outside the wall of the Grand Palace, Bangkok.
• Örebro slott May
• Giuseppe Arcimboldo – Summer

(Note: difficulties in locating facial parts)