NIST Special Publication 800-38A, Recommendation for Block Cipher Modes of Operation, states in Appendix C that (emphasis mine):
"For the CBC and CFB modes, the IVs must be unpredictable. In particular, for any given plaintext, it must not be possible to predict the IV that will be associated to the plaintext in advance of the generation of the IV. [...]
For the OFB mode, the IV need not be unpredictable, but it must be a nonce that is unique to each execution of the encryption operation. For example, the nonce may be a counter, as described in Appendix B, or a message number."
Appendix B also allows (and describes appropriate methods for) the use of deterministic initial counter blocks (they avoid the use of the term "IV") for the CTR mode.
SP 800-38A does, however, allow the generation of IVs for the CBC and CFB modes by encrypting a unique deterministic value:
"There are two recommended methods for generating unpredictable IVs. The first method is to apply the forward cipher function, under the same key that is used for the encryption of the plaintext, to a nonce. The nonce must be a data block that is unique to each execution of the encryption operation. For example, the nonce may be a counter, as described in Appendix B, or a message number. The second method is to generate a random data block using a FIPS-approved random number generator."
Ps. I'm a bit puzzled myself about NIST's advice against using predictable IVs with CFB mode, so I decided to ask a separate question about it.
Pps. In a paper published in 2008, "New proofs for old modes", Mark Wooding proves:
"that full-width CFB is secure if the IV is any ‘generalized counter’, and that both full-width and truncated $t$-bit CFB are secure if the IV is an encrypted counter."
Thus, CFB mode with full-block feedback is secure as long as the IVs are unique, while for CFB mode implemented with partial-block feedback using a shift register, the encrypted counter method recommended by NIST (which guarantees both uniqueness and unpredictability) should be used.
One reason for the difference is that CFB mode with low feedback lengths has some weak ("sliding") IVs, notably including the all-zero IV, that should be avoided. These weak IVs are rare, however, so a randomly chosen IV is extremely unlikely to hit them.