We analyze the partition function of three-dimensional quantum gravity on the twisted solid tours and the ensuing dual field theory. The setting is that of a non-perturbative model of three dimensional quantum gravity-the Ponzano-Regge model, that we briefly review in a self-contained manner-which can be used to compute quasi-local amplitudes for its boundary states. In this second paper of the series, we choose a particular class of boundary spin-network states which impose Gibbons-Hawking-York boundary conditions to the partition function. The peculiarity of these states is to encode a two-dimensional quantum geometry peaked around a classical quadrangulation of the finite toroidal boundary. Thanks to the topological properties of three-dimensional gravity, the theory easily projects onto the boundary while crucially still keeping track of the topological properties of the bulk. This produces, at the non-perturbative level, a specific non-linear sigma-model on the boundary, akin to a Wess-Zumino-Novikov-Witten model, whose classical equations of motion can be used to reconstruct different bulk geometries: the expected classical one is accompanied by other "quantum" solutions. The classical regime of the sigma-model becomes reliable in the limit of large boundary spins, which coincides with the semiclassical limit of the boundary geometry. In a 1-loop approximation around the solutions to the classical equations of motion, we recover (with corrections due to the non-classical bulk geometries) results obtained in the past via perturbative quantum General Relativity and through the study of characters of the BMS 3 group. The exposition is meant to be completely 11 This pair of variables is the one identified by Kapovich and Millson [77]. 12 Indeed, in the 3-dimensional context in which they were originally developed, such intertwiners are interpreted as quantum tetrahedra in the four-valent case m = 4, and more generally as quantum polyhedra for higher valencies [80][81][82][83]. See also [78,84] for a discussion of polyhedra in homogeneously curved space. In the context of canonical 3d gravity, the interpretation of SU(2) intertwiners as polygons was never truly developed beyond the interpretation of 3-valent intertwiners as quantum triangles. 13 If non-closing configuration are excluded, their contribution is suppressed in the large-spin limit [32], which admits in turn a semiclassical interpretation in terms of Regge geometries.14 Computing the expectation values of the su(2) generators on the state |j, j , we get J = (0, 0, j). In turn, we can compute the variance J 2 = j(j + 1), which is simply given by the su(2) Casimir. Thus the state |j, j corresponds to a semi-classical vector of length j in the z-direction, peaked on (0, 0, j) with spread 1 J J 2 − J 2 ∼ 1 √ j . The corresponding polar angle θ can be estimated to be θ ≈ arccos j √ j(j+1) ≈ 1 √ j → 0. 15 Indeed, the S2 in which n lives is better understood as SU(2)/U(1). This corresponds precisely to the Hopf fibration S2 ∼ = S3/S1.