This paper describes a technique for garment simulation and collision detection implemented on modern Graphics Processors (GPU). It exploits a mass-spring cloth model with velocity modification approach to overcome the super-elasticity. Our novel algorithms for cloth-body and cloth-cloth collision detection and response are based on image-space interference tests. For collision detection a 3D texture is generated, in which each slice represents depth and normal information for collision detection and response. These algorithms were implemented to build a fast web-based virtual try-on system. Our simulation starts from flat garment pattern meshes and performs the entire seaming and cloth draping simulation on the GPU. By mapping cloth properties of real fabric measurements we are able to approximate the real drape behaviour of different types of fabric, taking into account different warp and weft fabric drape properties. As the results section shows the average time of dressing a virtual body with a garment on state of the art graphics hardware is 0.2 seconds.