Bias are shot at the shortest exposure as possible, and with the camera capped, and at the gain that you use to capture light frames. Anything lower than 50 µs should be fine since there is hardly any dark current noise with such an exposure duration. The aim is to measure the sensor noise that has no dark current noise (mostly read noise) and where the camera A/D offset (often called "bias") is set to. Temperature is not critical, since temperature affects mainly the dark current noise.
Dark frames are captured at the same temperature, gain and exposure duration as your light frames, and also with no light entering the camera (be careful of light leaks from ZWO filter drawers and filter wheels). It captures the dark current noise and the read noise (your post processing program will know how to separate them), and also sensor defects (such as warm and hot pixels).
Flat frames are captured at the same temperature and gain as the dark frames, pointed at a defocused, uniform illumination whose spectrum covers the spectrum of your camera . The aim is to measure the precise light falloff (and vignetting) of the OTA. The exposure time is set so that the flat frames are as bright as possible, but does not saturate (i.e., average ADU as high as you can make it, without max ADU going over about 65000).
If you want to get good results, don't set ASIAIR to "auto" anything, including the exposure time. You can easily watch the ADU numbers yourself and produce more consistent results.
Now, you do not have to use the temperature of the "real" dark frames (for example, capturing the flat frames indoors), but in that case, you should also capture a set of dark frames (often called dark flats, or flat darks) at the same temperature as the flat frames. These dark flats are used to calibrate out the dark current noise and bad pixels in the flat frames. Good stacking and post processing programs will be able to handle these dark flats and not require you to use the temperature that you use for light frames.
Chen
