Spaces:

---

ltx-community

/

ltx-2.3-beard-removal

like 5

Running on Zero

Fuses the LoRA + loads the precompiled AOTI transformer blocks at the root module level (per ZeroGPU docs), using the STG-capable Group C repo (ltx-community/LTX-2.3-Transformer-GroupC-STG-sm120-cu130-rb3): the perturbation path is compiled as always-on tensor math, and a small per-block wrapper feeds a no-op ones mask when STG is off so one graph serves both. Keeps STG on (default stg_scale). Public graph-only repo → no token. Validated end-to-end on a day-to-night duplicate (coherent output, STG passes run clean).