trigger word is CBRPNK

trained on hand captioned videos. each video is a distinct video that is most likely a DAL-E 3 image with a custom prompt extended into a video with minimax on one platform or anther.

the training videos come from a home cyberpunk tabletop rpg game I run, so to me they have a somewhat consistent style, or I am willing to believe they all exist in the same world.

this was an experiement to see if hand labelling and loading videos without cuts helps.

Honestly, just taking the same prompt to a big model that hasn’t been fine tuned with a LoRA is generating more interesting results.