We describe a method for responsive, high-quality synthesis of human motion. Our method can quickly provide a motion synthesizer with a one-second long, high-quality transition from any frame in motion collection to any other frame in the collection. We construct these transitions using 2-, 3- and 4-way blends. During pre-processing, we search all possible blends between representative samples of motion obtained using clustering. The blends are evaluated automatically with a novel motion evaluation procedure, which we demonstrate is significantly more accurate than current alternatives. The best blending recipe for each pair of representatives is then cached. At run-time, we build a transition between motions by matching a future window of the source motion to a representative, matching the past of the target motion to a representative, and then applying the blend recipe recovered from the cache to source and target motion and whatever stubs are required. This method yields good-looking transitions between distinct motions with very low online cost.