I built CharaX to solve the dataset bottleneck for character LoRAs: getting images that are clearly the same identity across angles/poses (to avoid “average face” drift). You pick features → it generates a few candidates → you choose one → it exports a consistent 10-image multi-angle/pose pack (front/left/right/back + poses) for training. You could also upload your own chosen model and generate a consistent dataset from it. If you try it, I’d love feedback on: • does identity stay consistent across the grid (what drifted)? • are any angles/poses missing? • what export details would help (naming/captions/resolution)?