The initial conditions for, and the earliest stages of star formation are a very active research field. While many new observations in the mid-infrared as well as in the (sub-)millimeter range have produced a wealth of new data of star-forming regions in recent years, the observation-based knowledge about the earliest phases was still sketchy at best. The Herschel satellite provides unprecedented spatial resolution for the far-infrared wavelengths where such very young and embedded star-formation regions have the peak of their SEDs. We are conducting a Herschel Key Project (EPoS) to observe a selected sample of low- and high-mass regions with the PACS and SPIRE bolometer cameras. I will introduce our general Herschel programme and will mainly concentrate on the infrared dark cloud (IRDC) subsample. I will show that the embedded population of compact sources in IRDCs is traced now over a large wavelength range, and much more accurate SEDs can be constructed for these potential seeds of star formation, since identification ambiguities are finally resolved. Special emphasis will be put on the earliest manifestations of embedded entities: (a) unresolved emission sources only seen at Herschel wavelengths, but not with Spitzer/MIPS 24 micron, (b) high column density peaks within the IRDCs that still appear as extinction silhouettes even at 70 micron, indicating peak column densities >1023 cm-2. Combined with existing auxiliary data such as molecular line maps, we can assess the state of star formation for the embedded objects.