Transcoding is used in order to enable elegant adaptation of video sequences without fully decoding and recoding. For efficiency reasons, special attention is paid to transform-domain adaptation of video sequences, hereby avoiding the cost of forward and inverse DCT or, in the case of H.264/AVC, integer transforms. In this paper we compare different architectures for MPEG-2 to H.264/-AVC intra transcoding, both in pixel and transform domain. We provide a complexity and quality discussion of both architectures, and propose an alternative architecture which results in improved quality over the transform-domain approach at reduced computational complexity. We show that the complexity can be even further reduced by using a combination of different techniques. This results in the fastest transcoding architecture for all intra prediction modes while approaching the visual quality of the optimal rate-distortion curve.