Abstract : Freeform deformation techniques are powerful and flexible tools for interactive 3D shape editing. However, while interactivity is the key constraint for the usability of such tools, it cannot be maintained when the complexity of either the 3D model or the applied deformation exceeds a given workstation-dependent threshold. In this system paper, we solve this scalability problem by introducing a streaming system based on a sampling-reconstruction approach. First an efficient out-of-core adaptive simplification algorithm is performed in a pre-processing step, to quickly generate a simplified version of the model. The resulting model can then be submitted to arbitrary FFD tools, as its reduced size ensures interactive response. Second, a post-processing step performs a feature-preserving reconstruction of the deformation undergone by the simplified version, onto the original model. Both bracketing steps share streaming and point-based basis, making them fully scalable and compatible with point-clouds, non-manifold polygon soups and meshes. Our system also offers a generic out-of-core multi-scale layer to arbitrary FFD tools, since the two bracketing steps remain available for partial upsampling during the interactive session. As a result, arbitrarily large 3D models can be interactively edited with most FFD tools, opening the use and combination of advanced deformation metaphors to models ranging from million to billion samples. Our system also offers to work on models that fit in memory but exceed the capabilities of a given FFD tool.