fal(@fal):We trained a custom H3 Max LoRA on 335 clips of footage to generate scenes for an Emmy-finalist interactive murder mystery. fal creative engineer @PSDWizzard breaks down the full process, from story-engine JSON and character bibles to audio separation, captioning, and prompt compilation. A very good thread on what it actually takes to train a model for a specific creative world.

2026.09.16 20:51

We trained a custom H3 Max LoRA on 335 clips of footage to generate scenes for an Emmy-finalist interactive murder mystery. fal creative engineer @PSDWizzard breaks down the full process, from story-engine JSON and character bibles to audio separation, captioning, and prompt compilation. A very good thread on what it actually takes to train a model for a specific creative world.

Rabbit@PSDWizzard

2026.09.16 20:38

I tested 5,000, 8,000, and 10,000 step runs with the dataset. The captions mattered, adding an explicit matte toon style and descriptions of individual areas within each set helped anchor the look we wanted.