The main power use, and thus heat generator, are the camera themselves, they do use a lot of current when they are active. The ESP32S3 processor itself only uses circa 35mA when running and thats no really going to make it ‘hot’.
If you change the camera to an OV3660 you can put the ESP32S3 into light sleep and camera into standby and it uses circa 3.5mA.
Did anyone try to power the module directly with 3.3V (from a buck converter that is able to provide enough current) instead of 5V from USB?
I would expect, that the ESP32S3 attached with camera module can run significantly cooler because the onboard LDO converter from 5V → 3.3V no longer needs to convert voltage difference into heat.
Meanwhile I tried by myself by attaching an external 3.3V voltage regulator (a Pololu D36V6F3), feeding the 3.3V directly to the 3V3 pin of the ESP32S3 board.
The Pololu buck converter has a wide input voltage range of 4V … 50V. When I feed the buck converter Vin pin with 5V, it draws around 420 mA, when camera and WiFi are operating at full power to provide a continuous video stream.
Assuming an efficiency of the buck converter around 90%, this means, at 3.3V, the board draws around 580mA!
This means, the Pololu voltage regulator already operates at it’s limit and also gets very hot. The advantage now is, that the ESP32S3 board is running less hot than before, when the 5V → 3.3V reduction has to happen on the boards integrated LDO regulator (which has much less efficiency than the switching regulator).