Thank you @bresch for this quick answer!
So it would be the same for the Drag Coefficient (Cm)?
Then if I understand well from this topic , you only multiplied the Pseudo-Inverse Matrix (on the left side) of the matrix command (roll, pitch, yaw. thrust)? But the command are limited from [-1;1] for the Roll, Pitch, Yaw and from [0;1] for the thrust. Which gives you a normalise result.
After that you find a normalise PWM based on the command with a curve fitting from previous model?
Finally, you applied a scale factor base only from the parameter chosen, which is arbitrary to anyone. This value is then send to the rotor.
Thank you