I'm continuing my experiments with collecting datasets from larger neural networks. I've been remembering a game called Tribes: Vengeance for a while now, which featured a character named Julia, who belonged to the Imperial faction. I don't really remember much about the plot itself, but I couldn't find any LoRA models featuring this character, so I decided to try training one. I decided to try to "revive" it by attempting to collect a synthetic dataset in Nano Banana PRO and ChatGPT Image. I collected it quickly, with only 14 images. Overall, the result is definitely unstable in costume details. It would probably be more stable in dpmpp_2m_sde and sgm_uniform at 15 steps. But if you ever use this LoRA and find a more stable sampler and scheduler setup, let me know.

Triggers

jul1atribes

jul1asu1t

jul1atribes dressed in a jul1asu1t