Trained on the Anima base model. Also tested and confirmed working on Anima Turbo, using low-step settings (CFG 1, 8-10 steps) — so this LoRA isn't limited to full Anima inference, it holds up under Turbo's fast generation pipeline as well.